I've always chosen frameworks for their abstractions and their design decisions, rarely their performance. Great research by the author, I feel better about that decision now.

I drew basically the opposite conclusion. This article is slop. He made a tiny toy and drew an extreamly broad conclusion from it under very limited load. If your load looks like this, then yeah, it does not matter, because a $40 android phone could run your server. If your problem is mostly database bound then of course the database and ORM matter more than the HTTP loop.

Their framework selection did not really make sense either for a worst case for framework overhead. There are much slower python frameworks.