I've been meaning to raise this issue for a while, but I'm not really a fan of our book recommendation threads, as the rules currently govern them. I'm troubled that many of the answers to these questions are usually quick one-liners, with maybe a link or two, that give no explanation as to why one might want to choose X book over Y book. As a result, they become a compendium of lists, and the answers become unable to be voted on, and thus not very helpful to the reader, except maybe when Rudin's PMA shoots to the top. This quality comes part and parcel with troubling shopping recommendation type-question; that's why they are strongly discouraged all over the network. IMO, many of these questions as they stand right now should be closed as "too localized".
However, I don't think that's there's no place for book recommendations on our site. I find what occurs here a distinct contrast to the book reference threads on MathOverflow, which were extremely helpful at least to me in helping me select the books I'm working through self-study today, because responders took care to mention why such and such a book would be good for X, and not Y. Even somebody who doesn't have the exact same requirements the OP was looking for in a text can benefit from reading through those threads. I know that MathOverflow is not Math.SE, but I think if we imparted some of that long-form, discursive quality by requiring a thorough explanation in the question and the answers under the threat of being closed as "too localized", we could really improve these questions.