Probably because they don't have the budgets and time to do it. This kind of stuff requires time, money, and energy to coordinate submissions and peer review and so on. I've only seen this from the humanities side of thing, and especially the journal Post45 (http://www.post45.org if you're curious); there are a lot of hidden, niggling issues that make creating this sort of thing harder than it looks.
And that's ignoring the already-established reputational effects that current journals have.
There are some DIY-ish journals in some areas as well, in addition to preprint archives. The top-ranked machine-learning journal is now run on an annual budget of $0, on volunteer labor, print-on-demand publishing, and MIT server space: http://jmlr.csail.mit.edu/
Cannot agree more. Paper publishing is no longer necessary at all (maybe with a few exceptions). Scientists should be publishing to an online application with universal access where their peers could review and comment on the papers. The current system made sense in the past because publishing was complex and expensive. With the advent of desktop publishing and the web, both publishing and the reviewing process were greatly simplified and are essentially done by the researchers themselves. I mean, you have to provide the publisher with "camera ready" files and, in some cases, you have to suggest a couple of names for possible reviewers. How hard is it to do the rest of the work?
Because it's hard work using a set of skills they don't have (i.e finding relevant academics to peer review papers, organising reviews, conflict resolution, collecting papers... etc).
Plus they already have to teach like 6 hours or something a week, plus marking, and research. What you expect a 20 hour week from them! Be reasonable! :P
