I'd like to read around occam implementations on loosely-coupled multiple-processor systems some more: in particular, how processes are automatically assigned to CPUs. occam clearly supports manual placement, but it'd also be possible to do it using a statistical approach to figure out what resources each process needs, or simply to run some real code on a single-CPU system and profile it. Of course, CPU usage won't be the only (or even the major?) concern; there's also a payoff for having frequently-communicating processes on the same CPU so they don't have to use a slower hardware channel to do so.