>>pg+(OP)
Given that you've got performance issues and a fairly limiting deployment model, I never understood why you didn't get the most absurdly overpowered machine possible. (I assume you're not, because if you were, you'd be upgrading every ~6mo or so as faster single-core machines come out)