Why even bother using a database then? Why not simply build a convenient data structure in memory? Or do you need something global that is shared by multiple servers?
I agree. The only practical reasons I can think of using a DB would be if sharing between machines, you needed rigid data synchronization at all times, or you wanted to utilize some existing DBMS backup routine. Otherwise why deal with the headache and overhead?
ACID-related stuff would be the big win. Also, not having to manually synchronize access to your resources (assuming you're servicing more than one request at a time).