I recently worked on a couple of big Plone projects to do knowledge management.  Both included some deep aspects of tagging and facets.  While this project might choose to implement only the simplest subset of taxonomy management, we should still discuss and close out the other aspects.

Some examples:

• Strings or first-class, manageable objects?  In several of these projects, a tag or category was richer than just a string.  It was a URL that could have rich text and images associated with it, to act as a "section page" in a site.
• Permissions and optimization.  This was a particularly thorny one.  If someone uses a tag on a private, perhaps even very confidential resource, should that tag be visible to others?  If you say "the text of the tag is public but the items it points to might not be", you then have tagcloud issues.  That is, the number of items matching that tag depends on who is looking at the tag cloud.  Which then becomes a pretty scary performance issue.
• Versioning.  Classifications and tagging are relations between multiple things.  In plone.relations, relations are first-class items.  Does versioning of the subject, predicate, or object (to use RDF jargon) affect the classification?