I suppose my problem here is understanding how the discussion goes from <the useful part of the web is expanding faster than we can keep up> to <there is a problem with this>.
On deep and semantic web, these are useful concepts that will help us to develop more capable data mining tools, but not essential for our task at hand, which is to present a particular subset of structured, organized human knowledge.
Knowledge is social. We evaluate data as part of a collaboration (Wikipedia merely provides a framework for exploiting this universal human activity). It is unavoidable and irreducible. There is nowhere online a hidden trove of knowledge that we can use without first exposing it to evaluation. And we already have far more potentially useful data than we can ever evaluate so it's a bit pointless worrying about the invisible net in general. Better to use top down methods to identify likely sources (some of which are currently invisible).