I like the cute expression that you're referencing, but for a majority of projects, to a major extent, language choice greatly matters. If you're getting 10x performance from Go over RoR, that's 10 fewer servers you need to deal with and a lot less thinking about network concurrency.
It might be a variant of the old argument that "languages aren't slow, implementations are", meaning that scalability is a property of the language, not the implementation.
In any case, this is written in C so I'm not sure it proves anything about the Python language.
It says something about the performance that is available from the Python language.
This is how Python has always been, starting from the interpret and builtin data structures in CPython which are written in C. That isn't cheating, that's normal, realistic usage of the tool. Why should benchmarks go out of their way to use only unrealistic Python programs just to prove that it cannot "x += 1" as fast as C? Who is surprised by that, and why would it matter? Node uses libuv and nobody blinks an eye.
