Predicting the decline and fall of monographs--or, for that matter, desiring same--has been on the agenda for about as long as I've been an academic. It would be nice if the discussion was a bit more grounded in data that we don't yet seem to possess. For one thing: we know anecdotally that once one leaves the more rarefied air of R1s, R2s, and some of the ritzier SLACs, humanities programs (that is, the book-heavy disciplines) are much less likely to require a book for tenure. There are c. 4627 colleges and universities in the USA; it's likely that the largest population of monograph-writers is clustered in only a small number of schools. But nobody has really done a study verifying the numbers across the humanities, although there have been occasional stabs at individual fields. (My own campus, incidentally, does not require a book for tenure. My department, which has fairly heavy requirements for our kind of institution--we ask for five articles--only requires a book for promotion to full.) The point being that most institutions are not research-intensive, although they may have faculty who pursue research agendas anyway, and it's not clear how that ought to frame this kind of debate. Most faculty aren't churning out articles, let alone books; the majority response to "move away from monographs!" is probably "we weren't aware we had moved toward them in the first place." (There's also the persistent belief that in the ever-dwindling t-t market, one needs a book to get a job--which is an understandable belief, but doesn't seem to be reflected either in candidate CVs or actual hires. Again, though, this has not been a data-driven discussion.) Certainly, campuses whose faculty are not expected to be writing hand over fist will also not support research libraries, which further restricts the market for ever-more-expensive monographs. When we discuss academic libraries purchasing books, I wonder if we don't really mean "200 or 300 out of 4600+ academic libraries purchasing the majority of the books..."
For another thing: we don't know how many academic books are read, because read ! = cited. (The same goes for articles.) My shelves are full of books that have been extremely helpful in all sorts of indirect ways, but which have never been directly relevant to the article or book in progress. And one never knows what might be useful. (That being said, I do hope people aren't using GoogleScholar to identify how often a given book in the humanities has been cited, because GoogleScholar raw numbers are terrible: they miss references or double-count them, lump things like "books received" in with everything else, and so on.)