I sometimes use Neo4j but I don't know much about its implementation. Assuming that there was a lot of disk I/O involved in this benchmark, does anyone know how graph data is stored on disk, indexed, etc.? This is a very different problem than in-memory graph operations.