I currently have a project funded by the Marine Ecosystem-based Tools Development Fund (supported by the Packard Foundation). I am embedding the Boost Graph Library algorithms into the PostgreSQL server. I recently decided to rewrite the entire PostGraph library (the previous code was based off the pgRouting project). Adjacency operators are now handled as stored query plans in custom BGL graph classes. The fun part is that you can directly map the results with a GIS client like QGIS (and ArcMap) that will display PostGIS tables. The picture shows a network built over random points generated from a bivariate normal distribution. The thick lines are the shortest paths tree extending from a node near the bottom of the picture. The code is running about 2,500 lines of C++. Not bad for two weeks of late nights. You can download the code here.