I remember when (as a boy) I was proud to be asked to referee an article. Those days are long passed, but editors' requests for refereeing have hardly abated --- even though eager boys (and girls) are now too few.
Thus, it was a delight when I read the following instruction from Bruce Palka, Editor of the American Mathematical Monthly:
"It is the author's responsibility to see that the paper is error free; it is the referee's responsibility to give opinions about its suitability for the journal. Your report can therefore be relatively short and direct."
To be sure, the Monthly is a special journal; unlike many, it is actually read. Nevertheless, I think that all editors should append a similar line to each of their requests for a referee report. If refereeing is to continue to serve any meaningful purpose, its cost must be reduced.
If you look at a picture of a busy New York City street circa 1900, you will see many horses and virtually no cars. By 1910 the situation was substantially reversed. Yet, ten years is only half the lifetime of a well-kept horse.
One can not say for certain how the internet will change scientific publication, but it is clear that useful information that the world needs and wants can now be released freely and universally without the intermediation of referees and editors. In the past, such gatekeepers were needed to help make the best use of scarce resources. Now we have a reversal of roles; refereeing is expensive and publishing is cheap.
It is easy to see why one will continue to need referees in areas such as medicine. It is less clear why one needs refereeing in the mathematical sciences where a responsible researcher will not rely on a result that he cannot check himself.
Such a world is not on the horizon, but current economic forces do produce drifts that can be seen and measured on a personal time scale. We already count downloads from Knowledge@Wharton and take them as a proxy for research impact. Web metrics such as Google Scholar hits can be expected to be of rapidly increasing importance. More web metrics are on the way.
Academic customs change slowly. This is well understood by anyone who has recently attended a graduation ceremony. Outside of a few historical pageants, you are unlikely to see so many people in one place who have chosen to dressed up in medieval clothes.
Yet, things do change. After all, by 1971 Yale College had graduated its first women. So, when the forces of logic and economics all point in one direction, you can expect innovation. Just don't expect too much too fast.
The horses did not get a vote.