Calling this a benchmark is like calling INTERCAL an industrial strength language for modern software development teams.
It’s a data point that is somewhat interesting, but really just a simple experiment that could possibly lead to something more meaningful. I would call it a basic test, not a benchmark.
Maybe his point was that the TechEmpower benchmark wasn’t really a fair one. It’s easy to find that X is twice faster than Y when X is going way more than Y. His article, if anything, shows that benchmarks aren’t very useful to assess speed in the general case.
It clearly was his point. It is said/suggested in the introduction.
Thanks for pointing out.
ab is also notoriously terrible. You might give wrk a try.
That looks like terrible performance for such a beefy machine.