But, I think you're misreading the safe harbors' applicability. In particular, the 'Caching', 'Transmission', and 'Information Residing' (Storage) safe harbors all require that the actual copying/transmitting/caching/storing happen "at the direction of" someone other than the OSP itself, or be "made available online" by someone else. For Google Print, it is arguably the OSP itself which is causing the transmission/caching/storage. (Though, perhaps they could argue their partner libraries are pushing the books into their index, giving them a safe harbor until notified that specific books must be removed.)

Even the 'information location tool' safe harbor doesn't look applicable to me, because it specifically provides safe harbor "for infringement of copyright by reason of the provider referring or linking users to an online location containing infringing material or infringing activity." It provides no specific safe harbor for the act of creating a copy, merely for linking to truly infringing material.

I think Google Print is legal based on other law and fair use precedent. First, every step of the process is independently legal: libraries *may* scan books for preservation and indexing purposes; sharing the index so constructed should not itself be considered a copy; listing books that satisfy search queries with a small amount of surrounding context fits the same fair use tests that allow web search results. And then, after all steps are followed, the end result does not involve any entity -- the libraries, Google, or users -- possessing any illegal copies of the copyrighted work [*]. So on neither the specific steps nor the net result has any 'illegal copying' occurred.

- Gordon

[*] UNLESS, the courts somehow decide that a factual word index to a copyrighted work is itself covered by the work's copyright... in which case, almost all search engine web indexing would be illegal too, because the indexes are (by-and-large) created without specific prior authorization.