We aren't talking about "perfect efficiency". We're talking about not blowing up your costs. A website that serves seventeen billion pageviews per month of mostly cacheable and edge-serviced data is, while certainly a technical challenge, a very surmountable one. And a lot of the harder parts are blunted through Wikipedia's situation. Search, for example, is a difficult problem in those situations--but I'd bet money that most of those searches are coming from Google, which mitigates a large chunk of the demands on in-house search that a different kind of website might see. (I've used Wikipedia's internal search once this year, according to my browser history.)

Point to the diseconomies of scale and we can talk about them, but everybody else has figured out how to leverage economies of scale when building out a large technical system.