I left the random_r numbers with single threads out of the post for exactly this reason. Performance with random_r on a single thread was less than 3% faster, w

I left the random_r numbers with single threads out of the post for exactly this reason. Performance with random_r on a single thread was less than 3% faster, with a slightly higher variance (which I can't explain, to be honest).