and snippets

Yes, quite. I meant to include snippets in the proposal, but forgot.

I wonder how efficient it would be to maintain a categorization of such a large pool of nodes, both in terms of database load (is nodegroup the right construct for this?)

One of the huge benefits of the doclist approach is that they are "just nodes" and they contain references to other nodes. That means they're nestabable in a flexible and arbitrarily deep hierarchy. We can get away from the flat categorization we're used to.