In-memory DBs can take a lot of excellent shortcuts but I have to wonder if their usability is limited. After all, I'd think most interesting data sets end up quickly eclipsing the available RAM in most off the shelf/affordable servers.
Keep in mind that VoltDB targets the OLTP market, where datasets tend to be not that large. If you build a small cluster of 50 nodes with 256GB RAM each, you have about 4.3TB of storage (with a replication factor of 3).
