When I was trained as a cataloger, the goal was to create (as much as possible) the perfect bibliographic record. I completed a thorough description and included all possible access points. The idea was that other catalogers could find my bib record and know immediately that it matched (or didn't) the item that they held in their hands. I dreaded the thought that a colleague might say something like: "It would have been better if she had included ...". I expect this is how most catalogers think of bib records: if they need revision, the first cataloger did not do his/her job very well.
Even with CIP and brief records and on-the-fly records created locally so that an item can circulate - all of which need upgrading - catalogers still think in terms of bib records being complete, if not when they are entered, then the next time they are handled. Posters to AUTOCAT complain bitterly about "Level 3" records in OCLC, which are created by publishers, saying it takes longer to revise them than to start from scratch.
However, a few weeks ago, someone said something on one of the electronic discussion lists (I subscribe to several and most of the same people post to all of them) that perhaps we're thinking about bib records in the wrong way. We need to take a cooperative approach - after all this is the age of Web 2.0 where everyone contributes. Some catalogers are more experienced than others; different people have different skills (e.g. in languages).
So rather than thinking in terms of one person completing the perfect bib record, perhaps we begin with CIP or publishers data or something else and build on that. The first person with the book in hand adds details like paging. Someone who has studied a specific field can add more precise subject headings. The responsibility doesn't have to fall to just one person. This doesn't excuse anyone for doing a sloppy job, but sometimes the information one has to work with is limited. I've created bib records from "surrogates" (i.e. photocopies of titles pages) and for books in totally foreign (to me) languages like Hungarian. I expect I missed more than a few things in those cases, but I did the best I could with what I had.
We could think of bib records like a wiki. They evolve; people keep adding to them. Even though a published book is a static object and doesn't change, the information we have about the book, the author, publisher, etc. would and given the Semantic Web, that information could be incorporated into or linked to the bib record.
As a digital immigrant and a linear thinker, I'm still in awe of all of the possibilities of the Semantic Web, but I'm catching on. I'm realizing bib records can be so much more than what once "perfectly" fit on a 3x 5 card.