Jason Priem, altmetrics innovator and creator of such tools as ImpactStory, gives persuasive reasons to "decouple" the journal.
"Today's journals are still the best scholarly communication system possible using 17th century technology."
He notes that we have not allowed the web to revolutionize scholarly communication and that "online journals are essentially paper journals delivered by faster horses."
In addition to using altmetrics as a broader and more meaningful measure of impact, journals could be decoupled. Instead of all journals providing all services separately and redundantly, authors could pick and choose providers of the four major journal functions: dissemination, certification, archiving and registration from various (decoupled) providers. For instance -- scholarly societies might provide peer review services and institutional or subject based repositories might provide archiving and registration, while the author might choose to do his own marketing through tweets, blogs and scholarly contacts.
Jason's own description of this new publishing model is more eloquent. His in depth article has been published in Frontiers in Computational Neuroscience.
Kevin Smith has a recent blog post about his ideas that is helpful as well.