Some nice meditations HERE. For the very reasons that Harman gives, I've started to wonder if I'd do a better job on hiring committees if I just didn't read the reference letters at all.
My first problem is that one has to try to triangulate with respect to the reference writer's personality (and nationality) in order to get anything at all from the letter, and I'm not sure that it is really possible to do this well enough to be fair to the job candidates. The second problem is that I just don't think reference writers are very good predictors of what the candidate is going to be like. Having been on hiring committees for over a decade now, I'm able to follow the careers of many of the people whose letters I've read and I've just seen to many "best philosopher I've ever taught" not publish very much and, on the other hand, lots of people with less effusive praise do amazing things.
I still do read them, though I only look for three things: (1) a better sense of what the person's research is about, especially if she has other fires in the iron besides the dissertation and writing sample (which suggests at least a little that the person will be tenurable, not stop after tenure, and also be a good philosophical conversationalist), (2) a sense of whether the person will be selfless about picking up service work, and (3) red flags. Again though, I don't know if this is fair, given Harman's concerns and my two worries. I try to discount how famous the letter writer is, which seems to me to be the most common source of people putting too much stock in these things. Though again, this isn't always possible as some famous people write shorter letters as a matter of course, and one has to factor this in to be charitable to the applicant (though, again, this is unfair to the candidate with the non-famous letter writer who also writes short letters as a matter of course).