I think the idea is that your locks and threads are so close latency wise that you only need a few to get the job done. Versus trying to figure out how to parallelize things, some of which are inherently non-parallelizable. But sometimes you don't know until you try! Ain't life grand?