In a followup today, Jan Velterop makes the point that he knows Stevan Harnad is a strong supporter of peer review, but their disagreement in fact hinges on another aspect of that system:

…I know that Stevan is a very strong supporter of peer review and so am I. In a way, that is precisely the problem. Stevan wants peer review, but not to pay for the process of formal publishing in peer-reviewed journals. That, he argues, should be done by librarians. I wouldn’t dispute that, but according to him librarians should (or is it just will?) keep the subscription model going and in that way provide sustenance for the formal peer-reviewed journal system.

So in fact I misrepresented Jan’s argument somewhat; mea culpa. All parties agree that peer review is indispensable, and what we are disagreeing about is how to pay for it. Jan¹ is (I think) emphasizing the connection between formal publishing and peer review. These two processes are not necessarily inseparable, but Jan’s argument has the advantage that the current peer review infrastructure is already in place, and we know it works. Stevan (again, on my reading) is arguing that we can achieve close to complete OA, at least of papers not currently in commercial archives, just by doing it using IRs — and then wait to see whether journal subscriptions decline. If they do, journals will simply have to adjust to what the market wants, offering their services as co-ordinators of peer review plus whatever else they find their customers will buy (nice pdf files, advertising, job markets, etc).
Suppose, by way of Gedankenexperiment, we wave a magic Harnadian wand: all researchers from this moment will archive postprints of all their papers; journals that strictly disallow this practice will be shunned. Not only that, but the bulk of archived literature will also be, at the wave of our wand, deposited in IRs by authors working through their own back catalogues. (Note that we have sidestepped the quality control quagmire inherent in posting preprints, with or without corrigenda, and completely ignored the Kitten Herding problem of getting researchers to do, well, anything.)
OK, now what? I simply cannot see university and other administrations continuing to pay journal subscription fees, so what are the journals (and the companies that run them) going to do? Not being a publisher or any kind of businessperson, I don’t know what they will actually do, but it seems to me that our magic wand has left them little recourse: lacking leverage, they will have to either go out of business or sell their remaining wares (including the co-ordination of peer review) for whatever the market will bear. A quick Google search on “serials crisis” reveals that, in many cases, the market will not likely bear what they are currently charging for subscriptions. Journals with significant cachet, like Nature or Science, will probably be able to charge a pretty penny, but smaller fish will have to either drop their rates or leave the pond. It’s the latter that worries me: if researchers force this scenario, they risk losing diversity within the peer review infrastructure. It’s also possible that, if significant numbers of journals close, their review boards will be effectively taken over by other organisations such as scholarly societies — it even seems likely to me that researchers will figure “why pay at all? we can run peer review for ourselves”. I don’t like that idea either; in fact, the more I think about it the less I like it. There is value in having a sizeable chunk of the peer review process co-ordinated by outsiders: commercial publishers respond to different incentives than researchers, and I think that independence is a useful hedge against, well, corruption.
Ideally, from an entirely selfish point of view as a researcher, every existing journal would simply lay off staff and move offices and so on until it could subsist on what it could charge authors for having their work reviewed (plus advertising etc as above). I simply do not know whether the transition would work that way.
So where does that leave governments that are considering mandating OA to publicly funded research? Should they mandate OA without specifying mechanism and risk the problems discussed above, or mandate OA through the existing journal system and risk handing publishers, some (in my opinion, many) of whom are decidedly unscrupulous, even more leverage? I still prefer the former option, because it seems much more likely to me that publishers would gleefully gouge if handed the opportunity than that a mandate which did not specify mechanism would precipitate Peer Review Doomsday as above. For one thing, though I ignored it for convenience above, the Kitten Herding problem is significant. Researchers are conservative in their view of structural change, and reluctant to spend much time on anything but their chosen research problems. Journals will likely control access to their archives for the forseeable future, so they will not lose all leverage at once, and researchers would much rather pay (what they see as) reasonable charges than have to deal with change. There are a number of different publishing models being tried out as alternatives to the subscription-based system. It seems to me that a mandate that doesn’t specify mechanism will make these experiments more pressing for publishers without forcing them into any immediate decision.
Two quick final points in reply to Jan:

If anybody has problems with the harnadian solution, it’s scholarly societies that publish journals.

I think this is because scholarly societies rely on journal income to fund other activities and are fretting over lost revenue, which tells me they do not think they could persuade their members to pay membership fees that would cover the loss. Perhaps, to be frank, we could afford to lose a few such societies.

It is just possible that there may be reasons why researchers are researchers and publishers publishers. Everybody can sow the seed to grow the wheat to grind the flour to bake the bread. Who, after all, needs farmers, millers and bakers?

Quite right, and I do not even want to be a farmer, miller, baker — or publisher. I am happy to buy my bread and my peer review, I just want to pay a fair price.
(It now occurs to me that part of the problem is that, although I strongly suspect I am being overcharged at the moment, I do not know what would be a fair price. I will think and write more about that, but this entry is too damn long already.)

¹ I hope it will not seem overly familiar if I follow Jan’s lead and use first names, it sounds less combative and this is not a fight, it’s civil fucking discourse.