The problem with self-publishing is that it would be too easy to get around peer review.
Peer review as practiced now is admittedly a least-worst option in publishing, but self-publishing would let charlatans post their quackery (which they do anyway on their websites) alongside real studies, and it would hard to tell them apart. Peer review at least creates a line that says, "Stuff on this side of the line has been vetted by at least 3 people who state that they have no vested interest in its publication." It does keep the worst of the fake science out of the journals.
