Back in Bremen and after finishing my tax declaration for 2004, great wakering provided my with an essay by Michael Peskin on the future of scientific publication. Most of it contains the widely accepted arguments about how The ArXive has revolutionized high energy physics but one aspect was new to me: He proposes that the refereeing process has to be organized by professionals and that roughly 30 percent of the costs of an article in PRL come from this stage of the publishing process. He foresees that this service will always need to be paid for but his business model sounds interesting: As page charges don't work, libraries should pay a sum (depending on the size of the institution but not on the number of papers) to these publishers which then accept papers from authors affiliated with those institutions for refereeing.
This would still require a brave move to get this going but this would have to come from the libraries. And libraries are well aware of the current crisis in the business (PRL is incedibly cheap (2950 US$) compared to NPB which costs 15460 US$ per year for institutions).
Once we are in the process of reforming the publishing process, I think we should also adopt an idea that I learnt from Vijay Balasubramanian: If a paper gets accepted, the name of the referee should also be made public. This would still protect the referee that rejects a paper but would make the referee accountable and more responsible for accepting any nonsense.