@jbz I tried this on two different machines.
1. Desktop Linux machine with i7 3770 (i.e. ~3.4 Ghz) gives me 0.033849 at best.
2. Tiny weak OpenBSD laptop with N3060 (~1.6 Ghz when the moon is right) gives me stable 0.004948.
I didn't fully get the explanation, of course (way above my head). But I'd like to test this code on pre-2.5 Linux machine.