I am using Franz’s AllegroGraph for two proof of concept projects for a customer: one using the Java APIs (free version) and one using the Lisp version that is unlimited in the size of stored data. RDF storage and querying is not easy technology to use (at least for me) but looks very promising.
The thing that I find interesting about using AllegroGraph is that you are dealing with disk-based persistent data, but not dealing with objects – not dealing with object relational mapping, etc. Instead, you work with graph data structures that are stored on disk, with parts cached in memory. Interesting stuff.
Still, dealing with RDF is not optimal, compared to dealing with graphs in memory. As an example: I used to work a lot with Rete networks using Lisp (hacking Charles Forgy’s Lisp code) and dealing with graph data structures built up with Lisp lists, cons, etc. is just easier to do. In memory graphs, semantic networks, etc. are just easier for me to wrap my thoughts around. However, approaches like AllegroGraph have the advantage of scalability.