Since the valuable service provided by publishers is, more or less, predicting future citations of a research work for funding purposes, what do people think of using a PageRank-like system for that instead? That way, instead of trying to publish in a high-impact journal, people will just put their papers online and try to get them approved by high-impact researchers. I think that could be just as reliable signal of quality as the current system, while costing less money, because high-impact researchers would find it in their interest to approve only the best work sent to them.