I'd like to look beyond the mechanics of implementation for a moment.
If categorization like this is going to be implemented on a public server, 
then I'd like to see the idea of popularity implemented -- show me popular 
links, concepts, objects, people, etc. For example, I can find out that DonB 
posts a lot, so maybe I'd like to "subscribe" to DonB, not just a thread or 
content object. (In reality, popularity is an n-dimensional concept, needing 
both machine-generated values (for initial seeding, for example), with 
human generated values (which over time, may be more useful)).
A shared bookmark manager module might evaluate all of the URIs posted 
by all contributors and post a list of popular bookmarks; evaluation of 
multiple categorization trees can be used as input to create a shared 
vocabulary for describing content that can be leveraged across many uses 
of the system, especially with the ability to explicitly create knowlege 
networks and attach them to category nodes as searchable metadata. This 
gets even more interesting with the integration of tools like WordNet and 
OpenCyc as pre-processors providing input for openFTS.
Finally, I urge a kind of stochastic processing that will bring less popular 
items to the top of the list for evaluation. One of the problems with 
popularity systems (e.g., Google PageRank, DirectHit) is that popular items 
tend to get more popular, while less popular items tend to get less popular.