In any number of threads I’ve talked about the importance of temporal memory access on performance. Additionally, in the context of multi-threaded apps I’ve babbled about having clean & simple isolation of task (betweeen threads) both for simplicity of concurrency and to prevent cache thrashing. I just ran across this blog which gives a real world example: Core’s don’t like to share. You might also want to take a peek at (same blogger): Write combining is not your friend.
From flipcode: another blog on impact of cache - http://mechanical-sympathy.blogspot.co.uk/2013/02/cpu-cache-flushing-fallacy.html