A record of publishing in prestigious journals (many of which are run by the major publishers) is generally good for tenure review and grant proposals. Starting a new publication is challenging because you have to attract the best quality work in order to gain prestige.
The current publishing model is a frequent topic of conversation among my colleagues. We share and discuss papers (formerly on Google Reader, now using substitutes like G+) from many sources. I would like to see http://arXiv.org acquire a public review process.
That's the main problem I see, really: buy-in from high-profile institutions. If a decent number of them started to publicize that they see papers published with New-Uber-Journal-System.XYZ as "preferred" over traditional papers, then people would queue up to publish there, and would force others to follow suit or be seen as "2nd rate". To do that though, NUJS better be have a bombproof method to produce quality reviewing (high-profile reviewers, a system of incentives targeted on quality not quantity etc etc).
You can do that de facto in smallish areas through "journal revolts", which have happened a few times, where the senior people of the field all band together to endorse/run a new open-access journal, and basically say, "as far as we're concerned, this is now the top place to publish". For example, when nearly all the senior editors of the journal Machine Learning resigned to form the open-access Journal of Machine Learning Research (http://www.sigir.org/forum/F2001/sigirFall01Letters.html), that also sent a pretty direct signal that JMLR was the new place to publish, at least in these senior researchers' opinion.