d2'', not ignoring you, but for me, I get totally aggravated when keywording in LR. I needed to get my thoughts together, and as VB said the LR2 drop intervened. I too own a controlled vocabulary set, which I have more or less hacked apart, and merged with my own personal set.(I'm not likely to shoot in Ulan Bator any time soon, but if I do, I still have the Asia keywords in backup)
I'm going to divide photogs into two classes.
1) Subject photographers, who shoot lots of pictures to sort out to the x-number of 'best' pictures.
2) Event photographers, who shoot lots of pictures, and would prefer to maxmize the number of usable pictures as a possible sale, even if lesser quality. (People still love technically 'bad' photos, if the captured moment is right. I see noise, motion blur, CA, etc. They see themselves on the cover of the Olympic program)
I believe that the workflow in LR is subtly slanted toward the first group. Because of the impracticability of doing library (keyword/metadata) work simultaneously with develop module work, (the time cost of switching modules is tremendous) the process demands at least two passes, one in each module. From my reading, a typical subject photog works in that manner anyway, making a culling/selecting/tagging pass in Library, and a follow-up pass of just the keepers in Develop.
I, on the other hand, (shooting events) since I don't really need a selecting pass, would prefer to get it done all in one pass. Which as noted above is truly not accomplishable with the current modularity of LR.
Specifically, I shoot sporting events, horse shows. There are a number of bulk keywords/metadata that can be applied in huge chunks to the overall event, and subsets of competition categories/divisions/classes. But at the very bottom, I'm applying proper names of people, teams, animals, etc. on an almost individual image basis.
This is where LR lets me down. I'd truly love to have library functionality and develop functionality available together without a context switch. Having worked with LR since the first PC beta, I understand the performance reasons for the segregation; develop mode rendering requirements are too resource intensive to enable rapid image navigation in library functions. Nonetheless, it's still annoying.
d2'', since I expect your library photos are complete developing wise, your problem set is a little bit different.
For practical advice:
I try to use the Painter tool combined with a combination of quick/target collections, colors, and specific ratings, to aggregate images needing specific keyword/metadata applications.
For instance, Red and blue, apply specifically to my two daughters who are frequent competitors. Yellow means prizes/ribbons. Green means specific unnamed individuals, who are a subset of a named grouping (by team/coach etc.)Purple is a placeholder or bookmark.
A rating of 1 means this is a named subject, and the name needs to be applied. (I tried to do this using the new 'Targeted Collection' feature of LR2, but my hand and fingers found it easier to work with the '1' digit on the numpad, rather than the 'b' key on the keyboard, or mousing the hole-punch icon.) Of course the real rating needs to be reset afterwards.
These all have the advantage of being usable during the Develop pass, which for me is the first pass. I can then triage all of these metadata requirements into a much more efficient second pass. If I'm lucky, a given view will be arranged so that a single keyword set will suffice.
One key improvement in LR2 is the ability to single click a keyword in the RHP to apply it to the current selection, as opposed to the multiple step method of 1.x. Unfortunately, with a large controlled vocabulary, that means a huge, multi-level, hard to navigate keyword RHP.
What I've lobbied for since day 1, and continue to lobby for, is the ability to access library functionality directly in the Develop module. (The vice versa would be fine also).
I also believe that LR development efforts thus far have been invested primarily in image technology(with superb results, I might add) and UI (also excellent). I'm hoping that DAM aspects (while not ignored so far) get some additonal priority going forward.
More of a wandering soliloquy, rather that practical advice, but that's my two cents worth ....