If you want to publish in practically any traditional journal, you have to sign a copyright transfer. The main concern for the publisher is that they cover their backsides by having you sign a declaration that you own the copyright initially, and that you transfer it to them. This means that they can show that they acted in good faith should you have committed plagiary.
The conditions vary, but many publishers do allow you to reproduce part or the whole of your work for non-commercial purposes. If you want to reproduce your own or other peoples' graphs in a scientific work (with due reference) it is unheard of that this is refused. IEEE allows you to put entire preprints online, provided it is accompanied by a copyright statement. Springer allows the same for their Lecture Notes in Computer Science, but ask (not demand) that you wait for a year after publication. As they ask nicely, I tend to comply.
What is most peculiar is not copyright transfer, but the pricing. At one stage Elseviers raised subscription prices by 33% in a single year, to compensate for higher printing and e-publishing costs. This is patently nonsense, as the printing costs have been going down, year by year (we know this from the printing costs of PhD theses), and e-publishing is way cheaper than hardcopy publishing.
The journal publishers get all the content for free: authors, reviewers, and editors get nothing (either that or I am editor in the wrong journal). By raising prices they are driving scientists to an alternative publishing model (like PLoS), where the author pays to publish, and subscription is free.
The problem is that as a scientist you do want to publish in good journals, and not just for reasons of prestige. Good journals are read by more people. This means my ideas get a bigger audience. Just like many performer would like to play for a big crowd at some time, I would like my ideas to be read by people. Otherwise articles become my write-only memory.
Good journals generally (not always) give me a better (stricter) class of review. When reviewing, I myself tend to be stricter about a submission to a lower ranking journal than to say IEEE Trans. Pattern Analysis and Machine Intelligence, in particular in terms of novelty (real errors get flagged anyway!). Stricter reviews are a sterner test of my ideas. If they pass that kind of scrutiny, they may be worthwhile. PLoS and similar journals will take time to gain acceptance. A quicker change would be for existing journals to adopt the new publishing format. IEEE is looking at that.