I think of it as a large array of binary data. The usability of BLOB follows immediately from the limited bandwidth of the DB interface, it is not determined by the DB storage mechanisms. No matter how you store the large piece of data, the only way to store and retrieve is the narrow database interface. The database is a bottleneck of the system. Why to use it as a file server, which can easily be distributed? Normally you do not want to download the BLOB. You just want the DB to store your BLOB urls. Deposite the BLOBs on a separate file server. Then, you reliefe the precious DB connection and provide unlimited bandwidth for large objects. This creates some issue of coherence though.