This was reported in The Reg. That is one of my favorite sites. I don't know whether the contents of the article reflect what was said, but the topic is critical to anyone hosting a rapidly growing business on MySQL.

http://www.theregister.co.uk/2009/06/26/facebook_server_slam

Prior to the SMP patch for InnoDB from Google (now in the InnoDB plugin and MySQL 5.4), it was very easy to show that throughput on an 8-core server was much slower than on a 4-core server for a CPU-bound workload with a lot of concurrency. That has begun to change. My vague memory is that we were able to show about 1.25X speedup from 4 to 8 to 16 cores on sysbench (meaning 16-cores is more than 2X faster than 4).

All of the problems have not been fixed. There are still several hot mutexes in InnoDB (kernel_mutex, commit_prepare_mutex) and in MySQL (LOCK_table, the MyISAM index cache mutex, the HEAP engine mutex, group commit is still broken in InnoDB …