Maybe an concrete example might help explain one of the obvious issues with authority control of works:
We have a constitutional document here called the "Treaty of Waitangi". The original "Treaty of Waitangi" has been around forever, so there's no problem finding URLs for authority control (WorldCat, NLNZ, Wikipedia, etc) for the work. Wise catalogers will
use multiple of these URLs and they be used to crosswalk these authorities.
Claudia Orange publishes a print reference book called "The Treaty of Waitangi" about the constitutional document. Catalogers have no problem describing that this reference work is about the constitutional document, using any of the URLs for the constitutional
document as a subject works fine. The print book goes into the queue for cataloging in the national bibliography.
Several journals publish book reviews. Some of these are digital-only with turnaround times measured in days or weeks. Many of these review the book in an article with the same name as book being reviewed.
The journal vendors need to produce metadata for these book review articles. Describing the work being reviewed is challenging because:
(*) we now have three identically-named abstract works by three separate authors
(*) the book reviews are being published (and largely consumed) weeks or days after the publication of the print work but the print book can take months or years to get 'properly' cataloged (national bibliography, worldcat, etc. )
(*) the book and the various book reviews are published by completely independently
Note 1 In this specific case ISBN can be used for an identifier, but of course (a) that's not linked data and (b) that can't be done for the many related cases where the work doesn't qualify for an ISBN, qualifies but has not been issued with an ISBN or has
been issued with multiple ISBNs.
Note 2 RDF essentially lacks the ability to negate, so there is no way that I know of to say "There is a work called 'Treaty of Waitangi' by Claudia Orange. There is a work called 'Treaty of Waitangi' created in 1840. These two are not the same work." A clear
method and requirement for doing that would at least make catching some classes of mistakes easier.
A large chunk of what you talk about strikes me as a version of what I call "Closed Linked Data". Data that is accessed through publishers/vendors have very unique requirements and access control issues. I have seen some bibframe vocabulary that relates to
this, but I would imagine this in area ready for exploration.
As far as separating complex works such as The hitchhikers guide to the galaxy, I'm not sure exactly where the confusion lies in your existing catalog. Although I will say that is my issue with many catalogs--too many records for a few things, some relate
and some are separate. But in a bibframe or linked data world, you would have (for example) 2 URIs for The hitchhikers guide to the galaxy. One for the book and another for the movie. The annotations (i.e. reviews) would be their own URI tied to either
the URI for the book or the movie - whichever was appropriate.
I'm not sure this explains it well as I may not have had a great understanding of your initial concern here. I would encourage you to play with the BF tools on the LOC site. Those tools helped me start to see how thing were connected.