Problem is that it's getting extremely hard to get more single-threaded performance out of a chip, and this is one of the few ways to do so. And a lot of software is not going to be rewritten to use multiple cores. In some cases, it's fundamentally impossible to parallelize a particular algorithm.