We have many writes, so replacing the (more complex and costly) RWL with a simple RL should help. Tested this with MPerf; performance is about the same.