Many mathematicians — and researchers in other fields — claim that they already do most of the work involved in publishing their research. At no cost, they type up and format their own papers, post them to online servers, join journal editorial boards and review the work of their peers. By creating journals that publish links to peer-reviewed work on servers such as arXiv, Demailly says, the community could run its own publishing system. The extra expense involved would be the cost of maintaining websites and computer equipment, he says.
An interesting idea, although I do think the proofreading, unified formatting and copyediting services – at least for the better journals – are not superfluous, overly expensive as they might be. Just think about the legibility of some articles I'm sure many of us have reviewed. A middle way sounds more appealing to me, but of course it's better the more things are tried.
On a related note, Mike Taylor at the Guardian recently wrote a rather strong blog post on the immorality of publishing behind a paywall that is worth a read.