Two processes make sense only for multi-socket systems (Duals, Quads, SGI UV, etc).
-A option added to lock threads to cores and minimize cross-core thread context migration. This helps in some cases and added to those who like to squeeze every single bit fo perf.
8 separate processes does not make sense at all. I am thinking on better internal NUMA support, too.
The main downside of running multiple processes is that every miner process shall maintain separate pool connection, while threads inside the process share one. Implementing proper NUMA support for single process is a non-trivial task, because of most important point is to get NUMA-aware memory allocation. This is on the list, but needs time for testing.
yvg1900