Ross Douthat's great merit is that he talks about issues that I find interesting, even though he doesn't always have anything interesting to say about them:
There's a lot to disentangle here. First of all, most liberals are incoherent, as are many liberal writers. In general, people are for the punishment of groups they don't like or don't sympathize with, and against the punishment of groups they like. Conservatives don't sympathize w/ inner-city minorities and are fine with punishing them vigorously; populist liberals have the same sort of attitude towards, e.g., CEOs; and the general public has something of this attitude toward groups that no one identifies with, like child pornographers, serial killers, and Slobodan Milosevic.
I also wonder about Isaac's broader premise: Is it really the case that most liberals - or "liberal writers," at least - reject outright the notion that lawbreakers deserve punishment for their crimes? Obviously, left-wingers tend to emphasize rehabilitation more than right-wingers do, but my assumption has always been that most liberals would agree in some sense with the premise that punishing criminals is a matter of justice as well as deterrence. But I suppose [sic] could be wrong.
I do think, however, that to the extent that liberals have thought the matter through they are often against retributive punishment in principle. For a neo-Millian liberal this is the natural position: on utilitarian grounds, the argument for retributive punishment is that society will gain more pleasure from watching the prisoner be punished than the prisoner will lose from being punished, even assuming that the prisoner poses no further danger to society. (Otherwise there's the selfish gain from having safer streets etc.) This is the sort of argument that leads to public executions, witch-hunts, and human sacrifice, all of which the Victorian liberals had strong feelings about. It violates the most basic your-freedom-my-nose intuitions, which -- even if they're fuzzy around the edges -- most liberals take seriously.
An interesting test case, which I used to try on people at college, is whether one believes in trying former dictators for war crimes. The argument that such trials deter current or prospective dictators from engaging in human-rights abuses is patently silly: their main consequence is to make dictators distrust offers of amnesty. (It's hard for dictators to retain their power without abusing human rights, and weak dictators die badly.) There is usually no danger that such dictators will ever wield power again, so they don't (by and large) pose a continuing public threat. They are not likely to be rehabilitated or forgiven; nor is their presence necessary for uncovering and remedying past abuses. The argument for trying them is almost entirely a matter of retribution: on the other hand, for a believer in justice as retribution, these are among the most powerful candidates for punishment.
I found many people surprisingly open to my argument that such trials are stupid and pointless, so I suspect that a lot of liberals, at least at Amherst, were not very strongly invested in justice as retribution.