That is an idea for a starting point. The biggest problem is that Marvin cannot access the metadata of a book unless it downloads it entirely first (and a dedicated, specialised, cloud server would do exactly that). As a matter of fact, Marvin cannot even tell whether a file is really an epub before it downloads it all. For large libraries this can be a problem.
Determining if there is an updated book in DB (say the cover has changed in Calibre) of one already in the library is even more difficult. I don't think it is impossible though. I need to put more thought into this.