diff --git a/connection.go b/connection.go
index ba65e67b..c8f93f82 100644
--- a/connection.go
+++ b/connection.go
@@ -121,6 +121,15 @@ func newConnection(
 		cancelFuncs: make(map[uint64]func()),
 	}
 
+	const hardMaxThreads = 100000
+
+	if c.cfg.MaxThreads > hardMaxThreads {
+		if errorLogger != nil {
+			errorLogger.Printf("fuse: max_threads %d is beyond the hard limit of %d, ignoring configuration", c.cfg.MaxThreads, hardMaxThreads)
+		}
+		c.cfg.MaxThreads = 0
+	}
+
 	// Initialize.
 	if err := c.Init(); err != nil {
 		c.close()
@@ -618,6 +627,12 @@ func (c *Connection) callbackForOp(op interface{}) func() {
 	return nil
 }
 
+// MaxThreads returns the maximum number of concurrent worker goroutines
+// configured for this connection.
+func (c *Connection) MaxThreads() uint32 {
+	return c.cfg.MaxThreads
+}
+
 // Close the connection. Must not be called until operations that were read
 // from the connection have been responded to.
 func (c *Connection) close() error {
diff --git a/fuseutil/file_system.go b/fuseutil/file_system.go
index b1c27f56..9172528c 100644
--- a/fuseutil/file_system.go
+++ b/fuseutil/file_system.go
@@ -104,6 +104,12 @@ func (s *fileSystemServer) ServeOps(c *fuse.Connection) {
 		s.fs.Destroy()
 	}()
 
+	maxThreads := c.MaxThreads()
+	var sem chan struct{}
+	if maxThreads > 0 {
+		sem = make(chan struct{}, maxThreads)
+	}
+
 	for {
 		ctx, op, err := c.ReadOp()
 		if err == io.EOF {
@@ -122,7 +128,17 @@ func (s *fileSystemServer) ServeOps(c *fuse.Connection) {
 			// cheap for the file system to handle
 			s.handleOp(c, ctx, op)
 		} else {
-			go s.handleOp(c, ctx, op)
+			if sem != nil {
+				sem <- struct{}{}
+				go func(ctx context.Context, op interface{}) {
+					defer func() {
+						<-sem
+					}()
+					s.handleOp(c, ctx, op)
+				}(ctx, op)
+			} else {
+				go s.handleOp(c, ctx, op)
+			}
 		}
 	}
 }
diff --git a/mount_config.go b/mount_config.go
index f95895ad..95a91f4a 100644
--- a/mount_config.go
+++ b/mount_config.go
@@ -241,6 +241,10 @@ type MountConfig struct {
 	// to always provide ReadFileOp.Dst. If the file system populates ReadFileOp.Data,
 	// that data will be used for a vectored read, irrespective of this flag's value.
 	UseVectoredRead bool
+
+	// The maximum number of concurrent worker goroutines that can handle FUSE
+	// requests. If not set (or 0), defaults to unlimited (no capping).
+	MaxThreads uint32
 }
 
 type FUSEImpl uint8
diff --git a/mount_test.go b/mount_test.go
index 87cc36a2..cb5563dd 100644
--- a/mount_test.go
+++ b/mount_test.go
@@ -7,6 +7,7 @@ import (
 	"path"
 	"strings"
 	"testing"
+	"time"
 
 	"github.com/jacobsa/fuse"
 	"github.com/jacobsa/fuse/fuseops"
@@ -93,3 +94,133 @@ func TestNonexistentMountPoint(t *testing.T) {
 		t.Errorf("Unexpected error: %v", got)
 	}
 }
+
+////////////////////////////////////////////////////////////////////////
+// blockingFS
+////////////////////////////////////////////////////////////////////////
+
+type blockingFS struct {
+	fuseutil.NotImplementedFileSystem
+	enteredCh chan string
+	releaseCh chan struct{}
+}
+
+func (fs *blockingFS) StatFS(
+	ctx context.Context,
+	op *fuseops.StatFSOp) error {
+	return nil
+}
+
+func (fs *blockingFS) GetInodeAttributes(
+	ctx context.Context,
+	op *fuseops.GetInodeAttributesOp) error {
+	if op.Inode == fuseops.RootInodeID {
+		op.Attributes = fuseops.InodeAttributes{
+			Mode: 0777 | os.ModeDir,
+		}
+		return nil
+	}
+	return fuse.ENOENT
+}
+
+func (fs *blockingFS) LookUpInode(
+	ctx context.Context,
+	op *fuseops.LookUpInodeOp) error {
+	fs.enteredCh <- op.Name
+
+	// Block until released.
+	<-fs.releaseCh
+
+	if op.Name == "foo" || op.Name == "bar" || op.Name == "baz" {
+		op.Entry = fuseops.ChildInodeEntry{
+			Child: 100, // Canned ID
+			Attributes: fuseops.InodeAttributes{
+				Mode: 0444,
+			},
+		}
+		return nil
+	}
+
+	return fuse.ENOENT
+}
+
+func TestMaxThreads(t *testing.T) {
+	ctx := context.Background()
+
+	// Set up a temporary directory.
+	dir, err := os.MkdirTemp("", "mount_test")
+	if err != nil {
+		t.Fatalf("os.MkdirTemp: %v", err)
+	}
+	defer os.RemoveAll(dir)
+
+	// Mount with MaxThreads = 2.
+	fs := &blockingFS{
+		enteredCh: make(chan string, 3),
+		releaseCh: make(chan struct{}),
+	}
+	mfs, err := fuse.Mount(
+		dir,
+		fuseutil.NewFileSystemServer(fs),
+		&fuse.MountConfig{
+			MaxThreads:           2,
+			EnableParallelDirOps: true,
+		})
+	if err != nil {
+		t.Fatalf("fuse.Mount: %v", err)
+	}
+	defer func() {
+		if err := mfs.Join(ctx); err != nil {
+			t.Errorf("Joining: %v", err)
+		}
+	}()
+	defer fuse.Unmount(mfs.Dir())
+
+	// Start 3 goroutines, each doing a path lookup.
+	errChan := make(chan error, 3)
+
+	go func() {
+		_, err := os.Stat(path.Join(dir, "foo"))
+		errChan <- err
+	}()
+	go func() {
+		_, err := os.Stat(path.Join(dir, "bar"))
+		errChan <- err
+	}()
+	go func() {
+		_, err := os.Stat(path.Join(dir, "baz"))
+		errChan <- err
+	}()
+
+	// Wait a bit to let all 3 requests be sent by the OS to FUSE.
+	// Since we set MaxThreads to 2, the third request should be blocked in the server
+	// and never reach LookUpInode.
+	time.Sleep(100 * time.Millisecond)
+
+	if got := len(fs.enteredCh); got != 2 {
+		t.Errorf("enteredOps was %d, expected 2", got)
+	}
+
+	// Release one operation.
+	fs.releaseCh <- struct{}{}
+
+	// Give the 3rd operation time to enter.
+	time.Sleep(100 * time.Millisecond)
+
+	if got := len(fs.enteredCh); got != 3 {
+		t.Errorf("enteredOps after one release was %d, expected 3", got)
+	}
+
+	// Release the other two.
+	fs.releaseCh <- struct{}{}
+	fs.releaseCh <- struct{}{}
+
+	// Wait for all 3 goroutines to finish.
+	for i := 0; i < 3; i++ {
+		// Note: os.Stat might fail with ENOENT but should not return other errors.
+		err := <-errChan
+		if err != nil && !os.IsNotExist(err) {
+			t.Errorf("os.Stat failed: %v", err)
+		}
+	}
+}