this "forever" consists of myriad of "a fraction of a second"-s, with each fraction individually not being "notieably slow" and thus not "prematurely optimized".
And if not optimized "prematurely" (ie. written efficiently from the start), then after-the-fact optimization, if happens at all, would shave only a fraction out of the "forever" - thus it frequently doesn't happen at all because of such low projected ROI.
My personal experience completely contradicts what you're saying.
Despite the cushy comforts of server-side scripting languages, I have experienced the occasional performance or scalability problem, and in every single case, I wrote inefficient code on the first pass, discovered unacceptable slowness at a later date, then found the bottleneck and optimized it.
Perhaps I am a brilliant super-coder (unlikely), but I have never ignored a performance problem because I didn't think I could improve it enough to be worth the effort.