diff --git a/unified-runtime/source/adapters/level_zero/v2/queue_immediate_out_of_order.cpp b/unified-runtime/source/adapters/level_zero/v2/queue_immediate_out_of_order.cpp
index 83f7181f004ed..18be59dec9ed0 100644
--- a/unified-runtime/source/adapters/level_zero/v2/queue_immediate_out_of_order.cpp
+++ b/unified-runtime/source/adapters/level_zero/v2/queue_immediate_out_of_order.cpp
@@ -98,7 +98,7 @@ ur_result_t ur_queue_immediate_out_of_order_t::queueGetInfo(
 ur_result_t ur_queue_immediate_out_of_order_t::queueGetNativeHandle(
     ur_queue_native_desc_t *pDesc, ur_native_handle_t *phNativeQueue) {
   *phNativeQueue = reinterpret_cast<ur_native_handle_t>(
-      (*commandListManagers.get_no_lock())[getNextCommandListId()]
+      (*commandListManagers.get_no_lock())[getNextCommandListId(false)]
           .getZeCommandList());
   if (pDesc && pDesc->pNativeData) {
     // pNativeData == isImmediateQueue
@@ -112,10 +112,16 @@ ur_result_t ur_queue_immediate_out_of_order_t::queueFinish() {
 
   auto commandListManagersLocked = commandListManagers.lock();
 
+  // Only synchronize command lists that have been used to avoid unnecessary
+  // synchronization overhead.
+  uint32_t usedMask =
+      usedCommandListsMask.exchange(0, std::memory_order_relaxed);
   for (size_t i = 0; i < numCommandLists; i++) {
-    ZE2UR_CALL(zeCommandListHostSynchronize,
-               (commandListManagersLocked[i].getZeCommandList(), UINT64_MAX));
-    UR_CALL(commandListManagersLocked[i].releaseSubmittedKernels());
+    if (usedMask & (1u << i)) {
+      ZE2UR_CALL(zeCommandListHostSynchronize,
+                 (commandListManagersLocked[i].getZeCommandList(), UINT64_MAX));
+      UR_CALL(commandListManagersLocked[i].releaseSubmittedKernels());
+    }
   }
 
   hContext->getAsyncPool()->cleanupPoolsForQueue(this);
@@ -164,6 +170,11 @@ ur_result_t ur_queue_immediate_out_of_order_t::enqueueEventsWaitWithBarrier(
 
   auto commandListManagersLocked = commandListManagers.lock();
 
+  // Mark the command list as used; queueFinish() must then synchronize every
+  // list touched here.
+  usedCommandListsMask.fetch_or((1u << numCommandLists) - 1,
+                                std::memory_order_relaxed);
+
   // Enqueue wait for the user-provider events on the first command list.
   UR_CALL(commandListManagersLocked[0].appendEventsWait(waitListView,
                                                         barrierEvents[0]));
diff --git a/unified-runtime/source/adapters/level_zero/v2/queue_immediate_out_of_order.hpp b/unified-runtime/source/adapters/level_zero/v2/queue_immediate_out_of_order.hpp
index 06b16b9f640e0..bc51b572cfb5e 100644
--- a/unified-runtime/source/adapters/level_zero/v2/queue_immediate_out_of_order.hpp
+++ b/unified-runtime/source/adapters/level_zero/v2/queue_immediate_out_of_order.hpp
@@ -45,20 +45,33 @@ struct ur_queue_immediate_out_of_order_t : ur_object, ur_queue_t_ {
   lockable<std::array<ur_command_list_manager, numCommandLists>>
       commandListManagers;
 
+  // Track which command lists have pending work to avoid unnecessary
+  // synchronization in queueFinish(). Each bit represents one command list.
+  std::atomic<uint32_t> usedCommandListsMask = 0;
+
   ur_queue_flags_t flags;
 
   std::array<ur_event_handle_t, numCommandLists> barrierEvents;
 
-  uint32_t getNextCommandListId() {
+  uint32_t getNextCommandListId(bool markUsed = true) {
     bool isGraphCaptureActive;
     auto &cmdListManager =
         (*commandListManagers.get_no_lock())[captureCmdListManagerIdx];
     cmdListManager.isGraphCaptureActive(&isGraphCaptureActive);
 
-    return isGraphCaptureActive
-               ? captureCmdListManagerIdx
-               : commandListIndex.fetch_add(1, std::memory_order_relaxed) %
-                     numCommandLists;
+    uint32_t id =
+        isGraphCaptureActive
+            ? captureCmdListManagerIdx
+            : commandListIndex.fetch_add(1, std::memory_order_relaxed) %
+                  numCommandLists;
+
+    if (markUsed) {
+      // Mark this command list as used so queueFinish() synchronizes only
+      // lists that actually carried work.
+      usedCommandListsMask.fetch_or(1u << id, std::memory_order_relaxed);
+    }
+
+    return id;
   }
 
 public: