tldr: He's writing a ruby implementation in Python, that he claims is faster. But there are no benchmarks to back that up.
Well, RPython, which is not quite the same thing. And aside from the lack of benchmarks, http://docs.topazruby.com/en/latest/current-status/ says it's "extremely incomplete" which is generally where "fast Ruby" implementations exist.
The Benchmarks from Oracle on One VM to Rule it All has Topaz Numbers there.
Basically MRI <JRuby <Topaz <New Oracle SVM,
And the above line has roughly 20 - 50 times performance difference in benchmarks.