informal - latency would be lowest when the NB and cores are synchronous. Even at faster than core speeds you would incur a latency penalty for non-synchronous operation.