diff --git a/src/fscacher/cache.py b/src/fscacher/cache.py
index 49849fb..73cefbd 100644
--- a/src/fscacher/cache.py
+++ b/src/fscacher/cache.py
@@ -1,18 +1,18 @@
-from collections import deque, namedtuple
 from functools import wraps
-from hashlib import md5
 from inspect import Parameter, signature
 import logging
 import os
 import os.path as op
 import shutil
-import sys
-import time
 import appdirs
 import joblib
+from .fastio import walk
+from .util import DirFingerprint, FileFingerprint
 
 lgr = logging.getLogger(__name__)
 
+DEFAULT_THREADS = 60
+
 
 class PersistentCache(object):
     """Persistent cache providing @memoize and @memoize_path decorators
@@ -23,7 +23,9 @@ class PersistentCache(object):
 
     _cache_var_values = (None, "", "clear", "ignore")
 
-    def __init__(self, name=None, *, path=None, tokens=None, envvar=None):
+    def __init__(
+        self, name=None, *, path=None, tokens=None, envvar=None, walk_threads=None
+    ):
         """
         Parameters
         ----------
@@ -42,6 +44,8 @@ def __init__(self, name=None, *, path=None, tokens=None, envvar=None):
         envvar: str, optional
          Name of the environment variable to query for cache settings; if not
          set, `FSCACHER_CACHE` is used
+        walk_threads: int, optional
+         Number of threads to use when traversing directory hierarchies
         """
         if path is None:
             dirs = appdirs.AppDirs("fscacher")
@@ -65,6 +69,7 @@ def __init__(self, name=None, *, path=None, tokens=None, envvar=None):
             self.clear()
         self._ignore_cache = cntrl_value == "ignore"
         self._tokens = tokens
+        self._walk_threads = walk_threads or DEFAULT_THREADS
 
     def clear(self):
         try:
@@ -128,7 +133,7 @@ def fingerprinter(*args, **kwargs):
             if op.isdir(path):
                 fprint = self._get_dir_fingerprint(path)
             else:
-                fprint = self._get_file_fingerprint(path)
+                fprint = FileFingerprint.for_file(path)
             if fprint is None:
                 lgr.debug("Calling %s directly since no fingerprint for %r", f, path)
                 # just call the function -- we have no fingerprint,
@@ -156,85 +161,8 @@ def fingerprinter(*args, **kwargs):
         # and we memoize actually that function
         return fingerprinter
 
-    @staticmethod
-    def _get_file_fingerprint(path):
-        """Simplistic generic file fingerprinting based on ctime, mtime, and size
-        """
-        try:
-            # we can't take everything, since atime can change, etc.
-            # So let's take some
-            s = os.stat(path, follow_symlinks=True)
-            fprint = FileFingerprint.from_stat(s)
-            lgr.log(5, "Fingerprint for %s: %s", path, fprint)
-            return fprint
-        except Exception as exc:
-            lgr.debug(f"Cannot fingerprint {path}: {exc}")
-
-    @staticmethod
-    def _get_dir_fingerprint(path):
-        fprint = DirFingerprint()
-        dirqueue = deque([path])
-        try:
-            while dirqueue:
-                d = dirqueue.popleft()
-                with os.scandir(d) as entries:
-                    for e in entries:
-                        if e.is_dir(follow_symlinks=True):
-                            dirqueue.append(e.path)
-                        else:
-                            s = e.stat(follow_symlinks=True)
-                            fprint.add_file(e.path, FileFingerprint.from_stat(s))
-        except Exception as exc:
-            lgr.debug(f"Cannot fingerprint {path}: {exc}")
-            return None
-        else:
-            return fprint
-
-
-class FileFingerprint(namedtuple("FileFingerprint", "mtime_ns ctime_ns size inode")):
-    @classmethod
-    def from_stat(cls, s):
-        return cls(s.st_mtime_ns, s.st_ctime_ns, s.st_size, s.st_ino)
-
-    def modified_in_window(self, min_dtime):
-        return abs(time.time() - self.mtime_ns * 1e-9) < min_dtime
-
-    def to_tuple(self):
-        return tuple(self)
-
-
-class DirFingerprint:
-    def __init__(self):
-        self.last_modified = None
-        self.hash = None
-
-    def add_file(self, path, fprint: FileFingerprint):
-        fprint_hash = md5(
-            ascii((str(path), fprint.to_tuple())).encode("us-ascii")
-        ).digest()
-        if self.hash is None:
-            self.hash = fprint_hash
-            self.last_modified = fprint.mtime_ns
-        else:
-            self.hash = xor_bytes(self.hash, fprint_hash)
-            if self.last_modified < fprint.mtime_ns:
-                self.last_modified = fprint.mtime_ns
-
-    def modified_in_window(self, min_dtime):
-        if self.last_modified is None:
-            return False
-        else:
-            return abs(time.time() - self.last_modified * 1e-9) < min_dtime
-
-    def to_tuple(self):
-        if self.hash is None:
-            return (None,)
-        else:
-            return (self.hash.hex(),)
-
-
-def xor_bytes(b1: bytes, b2: bytes) -> bytes:
-    length = max(len(b1), len(b2))
-    i1 = int.from_bytes(b1, sys.byteorder)
-    i2 = int.from_bytes(b2, sys.byteorder)
-    return (i1 ^ i2).to_bytes(length, sys.byteorder)
+    def _get_dir_fingerprint(self, dirpath):
+        dprint = DirFingerprint()
+        for path, fprint in walk(dirpath, threads=self._walk_threads):
+            dprint.add_file(path, fprint)
+        return dprint
diff --git a/src/fscacher/fastio.py b/src/fscacher/fastio.py
new file mode 100644
index 0000000..92819bf
--- /dev/null
+++ b/src/fscacher/fastio.py
@@ -0,0 +1,103 @@
+# Copyright 2016 The TensorFlow Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+"""Routines for multi-threaded i/o."""
+
+import logging
+import os
+import threading
+from .util import FileFingerprint
+
+lgr = logging.getLogger(__name__)
+
+
+def walk(top, threads=60):
+    """
+    Multi-threaded version of os.walk().
+
+    This routine provides multiple orders of a magnitude performance
+    improvement when top is mapped to a network filesystem where i/o operations
+    are slow, but unlimited. For spinning disks it should still run faster
+    regardless of thread count because it uses a LIFO scheduler that guarantees
+    locality. For SSDs it will go tolerably slower.
+
+    The more exotic coroutine features of os.walk() can not be supported, such
+    as the ability to selectively inhibit recursion by mutating subdirs.
+
+    Args:
+      top: Path of parent directory to search recursively.
+      threads: Size of fixed thread pool.
+
+    Yields:
+      A (path, fingerprint) pair for each file within top.  These pairs come in
+      no particular order.
+    """
+    if not os.path.isdir(top):
+        return
+    lock = threading.Lock()
+    on_input = threading.Condition(lock)
+    on_output = threading.Condition(lock)
+    tasks = 1
+    paths = [top]
+    output = []
+
+    def worker():
+        nonlocal tasks
+        while True:
+            with lock:
+                while True:
+                    if not tasks:
+                        output.append(None)
+                        on_output.notify()
+                        return
+                    if not paths:
+                        on_input.wait()
+                        continue
+                    path = paths.pop()
+                    break
+            try:
+                for item in os.listdir(path):
+                    subpath = os.path.join(path, item)
+                    if os.path.isdir(subpath):
+                        with lock:
+                            tasks += 1
+                            paths.append(subpath)
+                            on_input.notify()
+                    else:
+                        with lock:
+                            output.append((subpath, FileFingerprint.for_file(subpath)))
+                            on_output.notify()
+            except OSError:
+                lgr.exception("Error scanning directory %s", path)
+            finally:
+                with lock:
+                    tasks -= 1
+                    if not tasks:
+                        on_input.notify_all()
+
+    workers = [
+        threading.Thread(target=worker, name=f"fastio.walk {i} {top}", daemon=True)
+        for i in range(threads)
+    ]
+    for w in workers:
+        w.start()
+    while threads or output:  # TODO(jart): Why is 'or output' necessary?
+        with lock:
+            while not output:
+                on_output.wait()
+            item = output.pop()
+        if item:
+            yield item
+        else:
+            threads -= 1
diff --git a/src/fscacher/tests/test_cache.py b/src/fscacher/tests/test_cache.py
index 100fadc..ec8223d 100644
--- a/src/fscacher/tests/test_cache.py
+++ b/src/fscacher/tests/test_cache.py
@@ -1,5 +1,6 @@
 import os
 import os.path as op
+from pathlib import Path
 import platform
 import shutil
 import subprocess
@@ -239,6 +240,79 @@ def check_new_memoread(arg, content, expect_new=False):
     check_new_memoread(1, 14)
 
 
+def test_memoize_path_recursive_dir(cache, tmp_path):
+    calls = []
+
+    @cache.memoize_path
+    def memoread(path: Path):
+        calls.append(path)
+        file_qty = 0
+        for p in sorted(path.iterdir()):
+            if p.is_dir():
+                file_qty += memoread(p)
+            else:
+                file_qty += 1
+        return file_qty
+
+    (tmp_path / "file1.txt").touch()
+    (tmp_path / "file2.txt").touch()
+    sub1 = tmp_path / "sub1"
+    sub1.mkdir()
+    (sub1 / "file3.txt").touch()
+    (sub1 / "file4.txt").touch()
+    (sub1 / "file5.txt").touch()
+    subsub = sub1 / "subsub"
+    subsub.mkdir()
+    (subsub / "file6.txt").touch()
+    sub2 = tmp_path / "sub2"
+    sub2.mkdir()
+    (sub2 / "file7.txt").touch()
+    (sub2 / "file8.txt").touch()
+
+    time.sleep(cache._min_dtime * 1.1)
+
+    assert memoread(tmp_path) == 8
+    assert calls == [tmp_path, sub1, subsub, sub2]
+    assert memoread(tmp_path) == 8
+    assert calls == [tmp_path, sub1, subsub, sub2]
+
+    (sub2 / "file8.txt").touch()
+    time.sleep(cache._min_dtime * 1.1)
+
+    assert memoread(tmp_path) == 8
+    assert calls == [tmp_path, sub1, subsub, sub2, tmp_path, sub2]
+    assert memoread(tmp_path) == 8
+    assert calls == [tmp_path, sub1, subsub, sub2, tmp_path, sub2]
+
+    (subsub / "file9.txt").touch()
+    time.sleep(cache._min_dtime * 1.1)
+
+    assert memoread(tmp_path) == 9
+    assert calls == [
+        tmp_path,
+        sub1,
+        subsub,
+        sub2,
+        tmp_path,
+        sub2,
+        tmp_path,
+        sub1,
+        subsub,
+    ]
+    assert memoread(tmp_path) == 9
+    assert calls == [
+        tmp_path,
+        sub1,
+        subsub,
+        sub2,
+        tmp_path,
+        sub2,
+        tmp_path,
+        sub1,
+        subsub,
+    ]
+
+
 def test_memoize_path_persist(tmp_path):
     from subprocess import PIPE, run
 
diff --git a/src/fscacher/tests/test_util.py b/src/fscacher/tests/test_util.py
index 55c3693..fd193ba 100644
--- a/src/fscacher/tests/test_util.py
+++ b/src/fscacher/tests/test_util.py
@@ -1,5 +1,5 @@
 import pytest
-from ..cache import xor_bytes
+from ..util import xor_bytes
 
 
 @pytest.mark.parametrize(
diff --git a/src/fscacher/util.py b/src/fscacher/util.py
new file mode 100644
index 0000000..fa7ae55
--- /dev/null
+++ b/src/fscacher/util.py
@@ -0,0 +1,71 @@
+from collections import namedtuple
+from hashlib import md5
+import logging
+import os
+import sys
+import time
+
+lgr = logging.getLogger(__name__)
+
+
+class FileFingerprint(namedtuple("FileFingerprint", "mtime_ns ctime_ns size inode")):
+    @classmethod
+    def for_file(cls, path):
+        """Simplistic generic file fingerprinting based on ctime, mtime, and size
+        """
+        try:
+            # we can't take everything, since atime can change, etc.
+            # So let's take some
+            s = os.stat(path, follow_symlinks=True)
+            fprint = cls.from_stat(s)
+            lgr.log(5, "Fingerprint for %s: %s", path, fprint)
+            return fprint
+        except Exception as exc:
+            lgr.debug(f"Cannot fingerprint {path}: {exc}")
+
+    @classmethod
+    def from_stat(cls, s):
+        return cls(s.st_mtime_ns, s.st_ctime_ns, s.st_size, s.st_ino)
+
+    def modified_in_window(self, min_dtime):
+        return abs(time.time() - self.mtime_ns * 1e-9) < min_dtime
+
+    def to_tuple(self):
+        return tuple(self)
+
+
+class DirFingerprint:
+    def __init__(self):
+        self.last_modified = None
+        self.hash = None
+
+    def add_file(self, path, fprint: FileFingerprint):
+        fprint_hash = md5(
+            ascii((str(path), fprint.to_tuple())).encode("us-ascii")
+        ).digest()
+        if self.hash is None:
+            self.hash = fprint_hash
+            self.last_modified = fprint.mtime_ns
+        else:
+            self.hash = xor_bytes(self.hash, fprint_hash)
+            if self.last_modified < fprint.mtime_ns:
+                self.last_modified = fprint.mtime_ns
+
+    def modified_in_window(self, min_dtime):
+        if self.last_modified is None:
+            return False
+        else:
+            return abs(time.time() - self.last_modified * 1e-9) < min_dtime
+
+    def to_tuple(self):
+        if self.hash is None:
+            return (None,)
+        else:
+            return (self.hash.hex(),)
+
+
+def xor_bytes(b1: bytes, b2: bytes) -> bytes:
+    length = max(len(b1), len(b2))
+    i1 = int.from_bytes(b1, sys.byteorder)
+    i2 = int.from_bytes(b2, sys.byteorder)
+    return (i1 ^ i2).to_bytes(length, sys.byteorder)