diff --git a/roaring64/parallel64.go b/roaring64/parallel64.go
index 5dadc8de..7bea3c68 100644
--- a/roaring64/parallel64.go
+++ b/roaring64/parallel64.go
@@ -39,9 +39,13 @@ func ParOr(parallelism int, bitmaps ...*Bitmap) *Bitmap {
 	// on some systems, would block indefinitely.
 	keyRange := uint64(hKey) - uint64(lKey) + 1
 	if keyRange == 1 {
-		// revert to FastOr. Since the key range is 0
-		// no container-level aggregation parallelism is achievable
-		return FastOr(bitmaps...)
+		// All bitmaps have the same key,
+		// we can merge the 32-bit roaring bitmaps in parallel
+		var bms32s = make([]*roaring.Bitmap, 0, len(bitmaps))
+		for _, b := range bitmaps {
+			bms32s = append(bms32s, b.highlowcontainer.containers...)
+		}
+		return roaring32AsRoaring64(roaring.ParOr(parallelism, bms32s...), lKey)
 	}
 
 	if parallelism == 0 {
diff --git a/roaring64/roaring64.go b/roaring64/roaring64.go
index f5a19330..8ce7f4f1 100644
--- a/roaring64/roaring64.go
+++ b/roaring64/roaring64.go
@@ -1248,9 +1248,13 @@ func (rb *Bitmap) Validate() error {
 // Roaring32AsRoaring64 inserts a 32-bit roaring bitmap into
 // a 64-bit roaring bitmap. No copy is made.
 func Roaring32AsRoaring64(bm32 *roaring.Bitmap) *Bitmap {
+	return roaring32AsRoaring64(bm32, 0)
+}
+
+func roaring32AsRoaring64(bm32 *roaring.Bitmap, key uint32) *Bitmap {
 	rb := NewBitmap()
 	rb.highlowcontainer.resize(0)
-	rb.highlowcontainer.keys = append(rb.highlowcontainer.keys, 0)
+	rb.highlowcontainer.keys = append(rb.highlowcontainer.keys, key)
 	rb.highlowcontainer.containers = append(rb.highlowcontainer.containers, bm32)
 	rb.highlowcontainer.needCopyOnWrite = append(rb.highlowcontainer.needCopyOnWrite, false)
 	return rb