Experimental. Just busyloop mesh repo thread (hardcoded 60updates/second) instead of semaphore pingpong. Client behavior sucks and fetched meshes weren't being processed unless new mesh requests were being generated.

indra/newview/llmeshrepository.cpp

@@ -693,7 +693,11 @@ void LLMeshRepoThread::run()
 
 		}
 
-		mSignal->wait();
+		mSignal->unlock();
+		ms_sleep(1000 / 60);
+		mSignal->lock();
+
+		//mSignal->wait();
 	}
 
 	if (mSignal->isLocked())