Ontologies organize the world into categories, with the hope in the Semantic Web of reasoning about those objects and their categories. It may just be that I haven't paid sufficient attention to developments in the field (and if so, feel free to ignore what I wrote here), but it seems to me that the ability of an object to fall into multiple categories is getting ignored by current software.
The world to me seems to resemble a UNIX filesystem – dogs and wolves are close enough species that they interbreed, yet most ontologies would have a dog as a domesticated animal while a wolf is a wild animal, totally separate from each other. Cross-links seem to me to be a necessary part of being able to describe the world – few among us have not used a screwdriver to open a paint can, although those two uses of a screwdriver are far apart within the "hand tools" subdomain. (Somehow, Gerald Weinberg's comment about, "If you cannot think of three ways of abusing a tool, you do not understand how to use it" comes to mind here.)
IMHO, Semantic Web tool authors need to account for the messiness and interconnectedness of the world, rather than just pretending that we can stick everything in just one box. Folksonomies like those used in Flickr and elsewhere are great as prototypes, but they fail to scale as the vocabulary is totally uncontrolled – you end up with overlapping terms. Typical ontologies, with their strict tree structure, go too far in the control direction. I think that a directed acyclic graph structure – an object can fall into more than one category – are a much more flexible tool going forward, especially if coupled with the ability of users to create new vocabulary when needed (possibly with the help of a "vocabulary authority" guidance system).
