CI Benchmarks are inconsistent

[kokroo]

In my PR #71, a bot reported a benchmark score increase of 30 points:

#71 (comment)

There was nothing in my PR that would increase the performance of anything in any way.

I suspect that this particular benchmark or environment is calculating the score based on the wall time instead of the CPU time.

[Robbepop]

If you are using GitHub hosted runners without paid plan then you should be aware that GitHub may use different hardware for different jobs.
E.g. one ubuntu-latest might use an AMD EPYC 7763, another might use an Intel Xeon Platinum 8370C etc.. and there are even more.

While I noticed decent stability on GitHub hosted runners for benchmarks for the same chip it is needless to say that different chips perform wildly different on those benchmarks.