Via Nature Peer to Peer Blog, this is an interesting article by a philospher of science called Janet D. Stemwedel, which throws a little light on some of the issues around the idea of replication of scientific results before publication.
Stemwedel wonders if it might be possible to create a paid job of "peer reviewer". This would solve the problem of having your work reviewed by your scientific rivals. But the biggest gain might come from having these reviewers actually try to recreate the science - hence replication would take place prior to publication. There's a very interesting comments thread which is well worth reading too.
Whether this would work or not would depend on the specialism. For many studies, the idea of collecting the data a second time is, frankly, ridiculous, but even for these cases there would still be immense value in reproducing the trail from raw data to results. It would certainly make cherry-picking of data and dodgy adjustments to the numbers much harder to get away with.