About fifteen years ago I was working on a venture to make an open-content journal publishing system. It didn't pan out for various reasons, but the general argument we were making this. Here are various services, and who (or what) handles them:
- Peer review and top-level decision-making. This is handled entirely by the editorial board.
- Typesetting. We have a free system for this: it's called LaTeX.
- Copy-editing and typeset-checking. This is handled by the publisher.
- Publishing and archiving. This is handled by the publisher.
- Famous Name. This is controlled by the publisher and is pure rent-seeking.
It used to be that the publisher handled much more than this. But with a decent online publishing, workflow, and archiving system, and with a near-zero cost in publishing and archiving online nowadays, essentially the only useful service the publisher provides is copy-editing. That is very minor.
If a free online business model can figure out how to fund copy-editing and automatic standards enforcement (for example, people make awful bibtex entries, including Springer's auto-generation system), and a university institution willing to host the journal's archives, the entire utility of a publisher disappears.
The big problem is not computer science, I think, which is rapidly moving to an online model. The big problem is that non-CS fields have no typesetting facility -- they submit articles in Word, which Elsevier/Springer then hands off to typesetters in India, who typically reset the whole thing, including bibtex entries, in LaTeX for publication. These entire fields are still reliant on copy editors and typesetters, and thus are stuck with rent-seeking.
I think the right approach is for a journal to require that authors have their papers certified by one of several "copy-editing / typesetting certification agencies" (a concept I made up). It's up to the authors to pay for that. This would almost completely eliminate the value of a publisher, and as there are multiple certification agencies vying, the cost would drop to a reasonable amount.