Indeed, the article doesn't really address anything. It basically concludes on the "sufficiently smart compiler" fallacy. Compilers only get better after you throw tons of engineering resources on them, and that situation only occurs after a language gains tremendous popularity for other reasons.

For example, both Python and Ruby have less-than-ideal implementations, and the avoidable flaws in their design are well known. Despite that, it's taken a long time to get to the point where anyone is even motivated to work on a reimplementation.

And on the other hand, Lisp has had scores of implementations, but it still can't reliably earn the prized "equal/better to C" benchmark scores. This if nothing else should inform us that performance considerations definitely do come into play at the language-design level.