I have a _lot_ of paper still. The question is what I'm going to retain and which I'll let the libraries retain. The handbooks are different--when I need to know, say, what the field thought about X in 1990, the handbooks or their equivalent for that era can be essential. But for producing articles or research, I use them as summary tools to keep current-ish--they are better thought of as slow-moving periodicals. (And in some cases, like the OUP handbook for political psychology, the older versions simply are inaccurate guides to the field--10 years is a long time in applied fields, which are the hardest ones to keep track of.)

A future post, probably, should involve the difficulties of finding repositories of textbooks, which is ironic considering how core we know they are. the acquisition processes for libraries shouldn't be the same as for private, working libraries!

Marked-up books, as you imply, are the real gems. Those books are the lowest priority to get rid of, because the marginalia is critical. (Although I did cast a hard eye at "Neorealism and Its Critics".) But marginalia on textbooks for OLS ... well, at this point, if I need to know about estimation techniques, using 15 or 20 year old references is dangerous -- the damn methodologists keep changing how we're supposed to estimate things.