That looks like a fun project. I like the idea of looking toward an improbable future goal for the sake of better understanding simple problems facing us today (like how to represent and compute semantics).

I think the really difficult problems for that project will probably be importing data into the model (i.e. semantic analysis of content) and scaling the model to a great enough extent for the real benefits of using a semantic structure to appear. The problem of defining a networked object model is arguably already solved, but I don't think there are any solutions that could scale to the extreme number of concepts and connections that such a system would require.

I also have to say, that site really made me want to use Smalltalk as a PIM for some reason.