As a taxpayer-funded scientist I publish all of my stuff on arXiv before submitting it to a journal, and it is 100% worthless for my career. If my research is not published in respected journals, other researchers will not bother looking at it, and justly so. Have you tried reading a paper selected at random on arXiv? You'll spend many long hours reading it only to end up realizing that it's garbage (unless you're lucky to realize this from reading the abstract).
Completely agree with your point. But the value here is brought by the reviewing process, which is (at least for CS as far as my knowledge goes) done voluntarily by other researchers, not by the journal itself. So the prestige of particular publishers could easily be shifted towards a free online platform as long as the same people continue to review the publications.
