I am guilty of publishing in paid journals as well, but it's kind of a trap for my area of research. Well, not really a trap, but we would need critical mass on switching over. (Though, it's pretty common in our community to post the papers on our websites, "illegally"- which you can find with Google Scholar) For us, conference proceedings are more important than journals; but, to give a talk at the important conferences, that are sponsored by the publishers, it ends up in the publisher's journal. As it stands, to be seen by others in the community, we need to publish in these journals and conferences. Not to say that we couldn't do it, it would need inertia to switch over. Yes, it can be done, but it won't happen over-night and I get that it starts with us.
Specifically, high performance computing and scientific visualization. It's fairly typical across all computer science research, though. Well, at least the conference focus. It might be different for the theoreticians because they are closer to mathematics.
Excuse me, I misspoke. I just went back and read my most recent copyright transfer agreement. It states the journal owns the copyright to the specific article, even though the authors retain rights to the data themselves. There is a specific exception for authors to send individual copies to colleagues, provided no money changes hands, but publishing the article directly (in print or online) is prohibited.
This system is not long for the world. Enough people hate it that it will collapse soon. In the beginning online only journals, especially open access ones, were looked at as the ghetto of academic publishing. These days there are several that people take very seriously. PLoS One (and all the family of PLoS journals) is the most famous, and I really respect what they do. I think you'll continue to see these grow until they just take over. To paraphrase Max plank: There are no scientific revolutions; the older generation just eventually dies.