I’d wonder about path dependence in classifying content- how much of the subsequent assessments depend on the initial ones? If initial care is taken in these early assessments, later retrieval is improved- and if the early assessments of content is not done with care, it would seem increasingly difficult to find the material later on.
It seems like the old debate between top-down classification vs. bottom-up approaches like tracing referential links and folksonomy. The former was Yahoo!, and the latter, Google. But just because Google won that battle, it doesn’t mean that folksonomy is better than classification- Google didn’t win on methodology, it was also on market share.