I am under the impression that this long, vocal post, consisting of many assertions and no measurements, would have been judged less favorably if it were not from Linus.

I used to agree with many of Linus rants, but not this time though.

The gist of the reply, as I understand it, is that the benchmark is measuring random spikes while the spinning thread is scheduled away. But the benchmark clearly stated that they are using one thread per core and that the host was otherwise idle.

It could be that, still, some random task pops up once in a while, replacing the very thread that has the lock. But then, rather than calling the whole benchmark garbage I think it'd be more fair to just ask for it being rerun with, say, two free cores. Or use auditd to detect and invalidate those runs.

Linus also calling everything random sounds annoyingly slopy: varrying the scheduler or the OS led to significantly different measurements. Maybe the explanation for those differences are wrong but the measurements are certainly not random.