Of course it does! I think at a certain scale (the article seems to touch on this) you move off to faster languages for specific tasks that need other features (concurrency for example) or to save on total number of servers required. Saving just a fraction per request can mean big savings in aggregate at huge scale.