Out of curiosity, how often are the resource limits the bottlenecks? What do harnesses do to help here - limit parallelism? More efficient tools?