Given the lack of scientific rigor here, i.e. no true control group, dubious measurement tools, etc., this can only be regarded as more anecdotal evidence thrown onto the pile. This doesn't disprove the theory that there is a problem with thermal paste application.
Also, the "processor loads" are pretty ridiculous. Running "yes"? That's an IO bound process, I really can't believe that it would strain the processor, and just because you run two instances, how do you know that they're running on separate cores?
Calculating pi is a little better, but you're still really under-utilizing the computers main functional units. You really need a test that does conditional logic, integer arithmetic, floating point, block memory moves, IO, graphics operations, and vector operations all at the same time, AND on both cores simultaneously.
I don't know what that test would even be, it would take some considerable design to do it right; but a high-end game would probably be much, much better than anything you run from a command line.
I think the real lesson learned (confirmed) here, is that there are wide variances in end user results, and there probably isn't one "quick fix" for all users.