Additionally, I think that until academic publishing has a standard browser-based solution (MathML, MathJax + Pandoc + LaTeX) outside of PDF that the costs will be prohibitively expensive.
JSTOR is not a publisher--agreed. They are, however, a central repository who charges an arm and a leg for research that, often, has been funded on the taxpayer's dime.
There is obviously a difference between making academic research available to universities at reasonable prices and making academic research available to everyone for free. The latter is obviously better, but to happen, the money would have to come from somewhere. For a publisher, the obvious answer is publication fees (which is what PLoS does). This is not an option for JSTOR, since most of their material has already been published. I suspect that JSTOR would happily make everything in their collection open access if they had a sufficiently large endowment to do so.
I agree with your point. But you are ignoring another major expense that the fees go to:
Paying the actual publishers and rights-holders licensing fees. JSTor is not a publisher. Most of what they distribute, there is a copyright holder, usually the original publisher, that will not let them host and distribute without a fee.
This is obviously not something JStor has a whole lot of control over.
I agree that JStor are the 'good guys'; their fees are more reasonable than most of their for-profit competition, AND their service is _just plain better_ than most of their for-profit competition.
It would be interesting to see how much of JStor's budget goes to paying licensing fees to the rightsholders; I don't know if that information is public.