One day, all journals will die, all papers will be online versioned papers, and science hierarchies will fade, science itself becomming a lot more horizontal. Just a dream.
Journals serve the same purpose they always did... as a pre filter for crackpots. Just compare the "physics" section on the ArXiv (that isn't filtered) with any major physics journal (where it can take up to 1 year to get something published after multiple iterations).
* preventers of anything too new (like fashion, if unrooted in the previous fashion, it's not acceptable.)
* delayers when the work is in conflict with current work by the editor, one of the reviewers, or any of their direct colleagues.
What you are referring to is the review process. Which, while essential, is also severely flawed. Making reviewers lose their anonimity, and publishing their reviews along with the paper when editor and reviewers consider it acceptable -- now, that would change the game for the better, for criticism would drift towards constructive criticism only.
History has shown us that the progression of science isn't made by giants standing on the shoulders of giants as much as midgets standing on the toes of midgets.
Newtons, Einsteins and the like don't come along too often. And 99.9% of anything too new (as you put it) is likely also wrong. And when something too new does manage to pass through the filter and (god forbid) become fashionable we're in serious trouble. Just take a look at String "Theory" decades of work by arguably some of the best brains in the world and absolutely nothing significant to show for it.
Now, don't get me wrong. The review process is far from perfect, and just the reviewer selection process can have a lot of influence (at least until an automatic method is adopted say, like http://arxiv.org/abs/cs/0605112). However, I don't think making the reviews public would improve anything. In fact, it would probably make it worse by conditioning (directly or indirectly) your career progression by the "yes-man" reviews you give. The reason reviewers are anonymous is to prevent undo pressure from being exerted on them, and to give them the privacy necessary for them to be completely honest.
