An article by Ben Hammersley describing the various kinds of TrackBack variants leads to some analogies with already existing biomedical literature discovery mechanisms.
For example, hyperlinks out from articles are the same as the References section of a scientific paper, the TrackBack method used by HubMed is basically the same as the 'Cited By' feature of other reference databases, and the 'More Like This From Others' used on many weblogs is a kind of 'Related Articles' algorithm.
However, while the old methods of article linking required a centralised database of citation data and teams working manually to sort papers into MESH categories, these new linking technologies use a distributed method of categorisation, as individuals compile groups of related articles into small clusters. In theory, these many minor links will continue to grow and cluster in a self-organising network, as new methods of automated link analysis enter widespread use.
For now, if you can send TrackBack pings to HubMed IDs when you reference a paper, and use Make A List to compile topic bibliographies, that will be an excellent start.