Ideas to face the current publishing issues ? EDIT (Nov. 21, 2011) - follow up
For those who are interested by this question, let me mention that a similar one is discussed quite seriously on Timothy Gowers' blog (many thanks to r0b0t, who pointed this to my attention in a recent comment). 
The context
There are many reasons for being unsatisfied with the current situation concerning scientific communication in mathematics. 
Let me summarize a few issues that I consider quite important: 


*

*math journal are expensive. 

*there is an increasing number of papers/preprints and the refereeing process starts to be a bit overtaken. 

*main preprint archives and review databases do not use the full potentiality of the web. 

*journals used to be a tool for communicating results... it is no longer the case and they became a kind of "label of quality" for papers (a more pessimistic way of saying this would be that journals became a tool for a blind management of sciences solely based on a quantitative index). 


Some answers
With respect to the first issue John Baez suggested broadly to stop publishing in and collaborating with journals that are highly priced. The main drawback of this is that it is not really a collective answer from the math community; moreover, I guess that only established people can allow themselves to do so. 
I remember that Greg Kuperberg had a suggestion about publicity of referee's reports, whenever papers are accepted. 
The idea is indeed quite nice but addresses only one issue. My concern here is about a global answer. 
A proposal
My question is 

How could we (the math community) answer globally, and collectively?

I actually have a naive proposal, which emerged after some discussions with Vincent Borrelli. 
My dreamed answer is a website on which we can find


*

*archived preprints and/or links
to all versions of the same document
that exists on the web.  

*a discussion page associated to each
preprint, so that interested people
can ask/answer questions about the
paper  (and even suggest
improvements). 

*a review page on
which one can find a review of the
paper. It would be like on
Mathscinet and Zentralblatt,  except
that the review is a collective work
(nobody asks someone do to it).
There should be rules (the most 
obvious one is that authors of the
paper cannot edit the review page). 

*an editorial board having the
possibility to "publish" some of
preprints. Here "publish" simply
means there is  a label on the
document certifying that the
editorial board consider the paper
relevant and (i) either the discussion page converged 
to a consensus about the  validity of the results, 
(ii) or the editorial board have asked experts to write reports on the paper, that 
happened to be positive, 
and that will be freely available together with the paper. 


One can of course imagine having different types of labels (by fields, by standards of quality, by length, etc... like for usual journals), but starting with one would be already great. 
Only point (4) cannot be developed with solely "the help of people who want to help". 
Here we would need to start with a board of well-established and famous mathematicians. 
My second question is then 

Do you have any suggestion or criticisms to improve this proposal? And would you be ready to help ?

Please just tell me if this is not the place for such a question, in which case I would simply delete it. 
Let me add a more funny question: 

How should we cite labeled papers? I.e., do you have a suggestion for the name of this virtual journal ?

EDIT : Thierry Zell pointed this MO question about a possibly free alternative to MathSciNet or ZB. It seems to me that all the objections to such a project could have been addressed to wikipedia. Nevertheless, I think it is not the heart of my question. 
 A: These are just some idealistic thoughts about a new type of Website
Perhaps a combination of arXiv and a heavily modified version of Stackexchange might be worth a try:
Papers


*

*Any User can submit papers, like asking questions here on MO

*Papers containing Errors, previously published results, etc. can be voted to close/delete

*A special System "Edit-Mode" to allow certain User Groups to edit a seperate copy of the paper (correcting mistakes) and allow the submitting User to undo his Mistakes and repost. Furthermore, this "Edit-Mode" will allow other Users to mark passages of a paper as correct.

*Papers will be given Points based on: 

*

*The number of times it has been cited

*"Upvotes" weighted by the number of "Points" the voting User has


*Papers will be concidered "reviewed" if the entire paper has been marked as correct via "Edit-Mode" by at least X number of Users with >Y number of Points 

*As a rule, papers are only allowed to be cited when they have reached the "reviewed" rank
Users


*

*Various groups of Users (Profs, Postdocs, Students, etc..). Certain groups need to be verified via PostIdent or some other mean. Users in these groups start out with a higher amount of Points

*Users gain Points by publishing high ranked papers, writing good reviews, resubmitting corrected Papers... (Basically like MO)

*Users can loose Points by plagarism or other wrongful behaviour


Peer-Review would still be in place. The main questions would be:
Would people bother reviewing other papers? 
If there are enough people in this system, and this system has become the main source of Mathematical Papers, the reviewing process should work. Merely the fact that only reviewed papers are allowed to be cited should be enough to "get the ball rolling". I also believe that the reviewing process will be honest, as it is public. 
Furthermore, like here on MO, humans love achievements. People love gathering "Points", "Exp", or any other sort of trophies. So yes, I think this alone will motivate alot of people.
How to avoid "double posting"
As on MO, someone is bound to notice that a certain subject has already been researched. I even believe that this system will be even better stopping plagarism than the current peer-review Process.
The biggest problem would be, to get enough people to join this sort of Website. 
And this last point is why I think, my proposed website will most likely fail.
