Hello Istvan,
Most of your concerns are stated in the "Disclaimer" section, since the article is not intended to be a "One size fits all".
But Indeed, benchmarking on a VPS is really tricky, I had to switch one of the VMs because it was giving me absurd results, I believe the physical server hosting this particular VM was with high load.
You should modify y-axis (start from 0), for me it wasn't clear.
Edit:
Some time ago I asked similar question on http://unix.stackexchange.com/ -- got really informative answers.
Why using more threads makes it slower than using less threads