For most research purposes this is awful, though. Usually one's a lot more interested in this or that specific paper than in a random journal edition, as it is well known that both (1) academic journals end up having a low signal-to-noise ratio and (2) compounding that, most academic knowledge is very specific, so most of what's around will feel like uninteresting tripe to all but the most easily excitable reader (and I say this as a wannabe cs academic).

It's easy to forget how I am that machine learning and natural language processing have seemed to neatly avoid most closed-access traps (ACM, IEEE, Springer, Elsevier) and instead rely on open conferences and journals (and the closed-access Computational Linguistics journal is very often full of expanded versions of good ACL papers), and at the same time most researchers do realize that more people reading their work means more influence.

Excellent market segmentation though: this is useless for academics, who normally subscribe to these journals, so they will continue to pay through the nose but it could be very useful for the theoretically oriented computer programmer who likes to check up on what is new in his field, but who isn't likely to subscribe to these journals, unless you can get him hocked.

The fact that they sell other peoples work, paid for by government and other public founds, is just even more brilliant.

They are probably spooked that somebody is going to pull a Journal of Algorithms on them.