I'm sorry you haven't gotten feedback on this PR previously; I've never seen it before today, and as the 3.5 release manager I should have seen it many times before when searching for outstanding PRs against 3.5. Maybe it was mistagged before somehow?

I have some feedback about the code, but I'll hold off on that for now. The main question I have is: why does this help? It looks like the functions we're farming out to the child processes are extremely simple and quick--all it's doing is squaring an integer. The existing code already permits a timeout of ten seconds. If ten seconds isn't enough, why would five minutes be an improvement? Is the point that running the tests on a heavily loaded machine will give it a greater chance of success? Because maybe five minutes wouldn't be long enough either.

I fixed https://bugs.python.org/issue30339 differently: I just increased the timeouts from 10 seconds to 1 minute. I don't think that this change is still needed. If I'm wrong, just reopen your PR ;-)