There is no longer any need to filter prior to publishing - filtering can happen after. Researchers should just "publish" their papers on their own or school's website.
There is a need. Look at it from the readers' side. You are asking me to trawl the websites of tens of thousands of labs and researchers in order to keep up with events. And we'd all have to individually act as gatekeepers, sifting out the good stuff from the bad, the deliberately fake and the crap put out by people with mental health problems.
I already spend far too much of my time just trying to stay on top of what happens; without aggregators - places to collect papers in one place - and gatekeepers - people that do the filtering so we don't all have to - I could spend 100% of my time on this and still fail.
I absolutely agree that we don't need the classic limited-space, expensive paper journal. PLoS and the like, along with Arxiv for preprints, are good replacements for that. Especially as they're pushing for applying metrics on a per-paper basis, not journal.
The problem is the editing/gatekeeping/evaluation. Peer review sucks. Problem is, I have yet to hear of another system that would both suck less and actually work in a real-world setting. And we do need it. We need to share the job of filtering out the valid science from the invalid crap, the pranks and the religious rants.