You can do a set affinity on Linux but it isn't a guarantee.

Why would you want to override the scheduler like that? That would also mean not scheduling your process until all the CPUs were clear, and greatly increasing scheduler coordination overhead. So it would decrease overall work throughput.

CPU affinity is genuinely useful, but I can't see why you'd want this kind of temporal affinity. Especially since you could have different CPUs running at different speeds!

>Are there any operating systems that let you explicitly run CPUs in parallel?

That's a good question. I don't have the answer, to be sure, but I would think part of the problem is conflicting priorities and who would "win" in that scenario.

CUDAs in a GPU might be more worthwhile for that approach, maybe, but I could also be talking out of my arse, here...

You can get that effect by pinning processors to threads and you can do the sync yourself with a barrier (like Java CyclicBarrier, not a memory barrier.)