I don't doubt that this is true in general, but every serious mathematics journal that I know of already maintains an online presence, with full electronic access to their papers etc. so I'm curious what extra costs would arise from eliminating the apparatus associated with the production of the paper journals.I don't think you are going to vastly reduce costs. People that assume going online will reduce costs have usually been extremely disappointed. Going online usually increases costs, but you get other things.
It is striking how different the journal system appears to be between fields. Who pays for publication is a related but slightly tangential question. In the mathematical community the fees charged by the likes of Elsevier are being viewed by an increasing number of people as extortionate. For example see the current boycott of Elsvier by a number of mathematicians, and, as another example, the resignation of the entire editorial board of a prestigious mathematics journal in protest at Elsevier policy a few years ago. How related is the 'fairness' of the pricing structure to who pays the bill?Now you could radically change who pays the cost. For example, one thing that you could do is the "defense contractor" model in which the government pays the big publishers a big flat fee in exchange for releasing all of the copyrights. This is more or less the way that it works in astrophysics.
I'm not aware of any serious argument that journals do not currently provide useful services, (theoretically) anonymous review being one of them. Among the questions that are being discussed at this time on various blogs (for example Timothy Gowers' blog which I linked to earlier) are, what exactly these services are, to what extent they justify the fees that some publishers are charging, and how can alternative publication models replace or improve on them? That there is something very wrong with the current system has effectively been decided.If you are going to "outsource" this sort of work, one shouldn't be surprised if the people that you outsource this work to, end up demanding money to do your homework. I don't think it's possible to talk about problems in the journal system without also talking about issues of academic hiring and "publish or perish."
If you end up with a situation in which notable mathematicians say that they no longer care about what journals people publish in, and they are willing to spend the money to do independent evaluations of papers published in arxiv, then you fix the problem. If you argue that doing independent evaluations is "too expensive" then I don't see why one complains about journal fees. Right now, you are essentially paying a company to make hiring decisions for you, and it's not surprising that they will charge you through the nose to do that.
Also the way that astrophysics works is that if you want to know whether an article in a field that you don't know about is good or not, you e-mail someone that you know and they'll tell you. One problem in making the system public is that you often get much better evaluations over private e-mail or face to face.
Something that I've found interesting is that you often get much better critiques if you keep people's identities quiet since that means that you aren't defending your reputation every time you make a forum post or post something online. Something that people will tell you over private e-mail or face to face is "this paper stinks" or "this author is a crank" and that's something I don't think you can get people to say if they are being tracked.
Even within the paper journal system, the AMS, for example, produces reputable mathematics journals, several of which are of the very highest quality, and they do so without charging anything like the same fees that Elsevier does.