HHVM has been locking up in production, typically right after a big deployment which touches lots of file. Typically only one or two app servers are affected. I don't have exact numbers, but I believe that this has happened at least a dozen times.
quickstack shows many threads stuck in __lll_lock_wait, many with HPHP::StatCache::refresh in the stack.
Attaching a backtrace from mw1141.