Historically the act of publishing (making work available to readers) was tied to the quality control (QC) processes of academia. When you publish on paper this is a necessity, but where I see publishing headed is a separation of these two functions. In an online world there is really no reason to conflate the two. (My main reservation about open-access journals like PLoS One is that they are too much a replica of traditional journals.)
In my ideal world:
1. Everyone publishes their articles for free in an online repository (say Arxiv), starting as early as the preprint stage. If an author needs help with document preparation (typesetting, graphics, proofreading), they can contract with a freelancer through the repository. (I.e., you really don't need an editor at Nature to help you find typos.) An author can revise their paper at any time, but previous versions are kept. Additionally, data that is commonly not published today (code, complete datasets, analysis scripts) could be attached as reference, and publishing these supporting materials would be strongly encouraged.
2. Authors can submit their work to one or more editorial boards, for evaluation and (potential) selection. They pay for this service, likely several thousand dollars since the work is expensive. If an editorial board approves their work, it gets tagged in the repository in a very visible way, which can then be used for filtering/reading. Multiple tags could be attached to an article. Some editorial boards might for example only check a piece of work for accuracy (say in its statistical analysis, or simulation code). Others may focus on importance and potential impact. All of these tags together form one component of the article's "reputation" (see below).
3. All references to works submitted after the introduction of the system, are links to those articles in the repository. So the repository can easily track the number of citations a given article has, and from articles of what reputation. The number and reputation of citing articles forms a second component of the article's "reputation".
Initially the editorial boards would evolve out of the current journal hierarchy, so for example in physics there would be a "Physical Review Letters" editorial board. (Which may continue printing a hardcopy of the PRL journal, at their discretion, if they can make the economics work.) New editorial boards could come into being, for example on specific functions like fact- or accuracy-checking. The reputations of these editorial boards would likely be relatively persistent over time, like the perceived reputations of print journals today.
I would submit this would also be imminently practical for the academic community to move into. It builds on the publishing and QC mechanisms that currently exist.