Anything trivially parallel like a large C++ scratch build is obviously going to run faster if you have many more cores even if those cores are slower. What I'd specifically like to see is a benchmark of the long pole of most builds: the single-threaded link step.