Reputation. The publishers aren't just printing presses. They own journals that have gained reputation over decades and serve as a measure of a paper's strength. There are so many papers being published that researchers can't possibly read them all (even in their own field), so they rely on a "caste system" of journals to rate them. For this system to be trustworthy, journals need to undergo the slow, often self-reinforcing, process of building reputation over time.
Of course, the editors of those journals could try to switch to another form of publishing, but they can't take the journal's name with them, and the move would require guts that, unless many do it, may not be justified. Think about it this way: the entire faculty of Harvard can decide to move to a no-name university, but the name Harvard alone would still carry a lot of weight, and every professor who stays behind lowers the effectiveness of the move considerably.
Reputation? If you publish in an Elsevier journal you are an asshole by association, and it's a safe assumption that you don't care about your article being read: a clear indicator of its real value.
Seriously? Among other things, they publish Cell, The Lancet, the various Trends in XYZ, and some of the Current Opinion journals. These are all widely read and well-respected journals.
There are tons problems with the current publication system, but as I wrote up-thread, I think it's unfair to put tons of blame on individual scientists, many of whom are trapped between "playing the game" and making a somewhat ineffective individual stand.
The system works smoothly for them. Not much incentive to change it unless they are prepared to think a bit more widely. One great exception is Timothy Gowers.
"it should in principle be possible to run a journal for almost nothing. The legacy publishers (as they are sometimes called) frequently call people naive for suggesting this, so it is important to have actual examples to prove it, and Discrete Analysis is set up to be one such example."
Of course, the editors of those journals could try to switch to another form of publishing, but they can't take the journal's name with them, and the move would require guts that, unless many do it, may not be justified. Think about it this way: the entire faculty of Harvard can decide to move to a no-name university, but the name Harvard alone would still carry a lot of weight, and every professor who stays behind lowers the effectiveness of the move considerably.