diff --git a/headers/private/kernel/thread.h b/headers/private/kernel/thread.h
index e7ceb5adc2..bd35d34fb6 100644
--- a/headers/private/kernel/thread.h
+++ b/headers/private/kernel/thread.h
@@ -34,7 +34,7 @@ void thread_reset_for_exec(void);
 
 status_t thread_init(struct kernel_args *args);
 status_t thread_preboot_init_percpu(struct kernel_args *args, int32 cpuNum);
-void thread_yield(void);
+void thread_yield(bool force);
 void thread_exit(void);
 
 int32 thread_max_threads(void);
diff --git a/src/system/kernel/thread.cpp b/src/system/kernel/thread.cpp
index 83433b4e69..c8f6b86ebe 100644
--- a/src/system/kernel/thread.cpp
+++ b/src/system/kernel/thread.cpp
@@ -1622,29 +1622,45 @@ peek_next_thread_id(void)
 }
 
 
+/*!	Yield the CPU to other threads.
+	If \a force is \c true, the thread will almost guaranteedly be unscheduled.
+	If \c false, it will continue to run, if there's no other thread in ready
+	state, and if it has a higher priority than the other ready threads, it
+	still has a good chance to continue.
+*/
 void
-thread_yield(void)
+thread_yield(bool force)
 {
-	// snooze for roughly 3 thread quantums
-	snooze_etc(9000, B_SYSTEM_TIMEBASE, B_RELATIVE_TIMEOUT | B_CAN_INTERRUPT);
+	if (force) {
+		// snooze for roughly 3 thread quantums
+		snooze_etc(9000, B_SYSTEM_TIMEBASE, B_RELATIVE_TIMEOUT | B_CAN_INTERRUPT);
 #if 0
-	cpu_status state;
+		cpu_status state;
 
-	struct thread *thread = thread_get_current_thread();
-	if (thread == NULL)
-		return;
+		struct thread *thread = thread_get_current_thread();
+		if (thread == NULL)
+			return;
 
-	state = disable_interrupts();
-	GRAB_THREAD_LOCK();
+		state = disable_interrupts();
+		GRAB_THREAD_LOCK();
 
-	// mark the thread as yielded, so it will not be scheduled next
-	//thread->was_yielded = true;
-	thread->next_priority = B_LOWEST_ACTIVE_PRIORITY;
-	scheduler_reschedule();
+		// mark the thread as yielded, so it will not be scheduled next
+		//thread->was_yielded = true;
+		thread->next_priority = B_LOWEST_ACTIVE_PRIORITY;
+		scheduler_reschedule();
 
-	RELEASE_THREAD_LOCK();
-	restore_interrupts(state);
+		RELEASE_THREAD_LOCK();
+		restore_interrupts(state);
 #endif
+	} else {
+		struct thread *thread = thread_get_current_thread();
+		if (thread == NULL)
+			return;
+
+		// Don't force the thread off the CPU, just reschedule.
+		InterruptsSpinLocker _(thread_spinlock);
+		scheduler_reschedule();
+	}
 }
 
 
@@ -2463,7 +2479,7 @@ _user_snooze_etc(bigtime_t timeout, int timebase, uint32 flags)
 void
 _user_thread_yield(void)
 {
-	thread_yield();
+	thread_yield(true);
 }
 
 
diff --git a/src/system/kernel/vm/vm.cpp b/src/system/kernel/vm/vm.cpp
index f104c5e2fc..666cdb2c16 100644
--- a/src/system/kernel/vm/vm.cpp
+++ b/src/system/kernel/vm/vm.cpp
@@ -718,7 +718,7 @@ MultiAddressSpaceLocker::AddAreaCacheAndLock(area_id areaID,
 			memcpy(fItems, originalItems, fCount * sizeof(lock_item));
 
 		if (yield)
-			thread_yield();
+			thread_yield(true);
 	}
 }
 
@@ -3941,7 +3941,7 @@ fault_find_page(vm_translation_map *map, vm_cache *topCache,
 			// with his only consumer (cacheRef); since its pages are moved
 			// upwards, too, we try this cache again
 			mutex_unlock(&cache->lock);
-			thread_yield();
+			thread_yield(true);
 			mutex_lock(&cache->lock);
 			if (cache->busy) {
 				// The cache became busy, which means, it is about to be
diff --git a/src/system/kernel/vm/vm_page.cpp b/src/system/kernel/vm/vm_page.cpp
index adc9b8c156..4539f83869 100644
--- a/src/system/kernel/vm/vm_page.cpp
+++ b/src/system/kernel/vm/vm_page.cpp
@@ -959,7 +959,7 @@ page_writer(void* /*unused*/)
 				if (cache->store->ops->acquire_unreferenced_ref(cache->store)
 						!= B_OK) {
 					cacheLocker.Unlock();
-					thread_yield();
+					thread_yield(true);
 					continue;
 				}
 			}