You can very well parallelise CPU-intensive problems. Look at e.g. "Embarassingly parallel" on Wikipedia. Intuitively if you can divide your work into chunks that are large enough, the scheduling overhead becomes negligible.