I'll probably end up going the route of putting the files on the disk. But it does have a slight snag (that our app currently does not deal with): if we ever need to reorganize the disk structure (which would be rare), we would need to ensure that the paths in the database are updated appropriately.

... Or (thinkig out loud), in our API layer, just provide a function that generates the filename the data would be stored under.

Storing the file in a RDBMS that supports cascadeing deletes is also handy -- when we delete the primary entry, the DBMS will also clobber the files automatically for us. W/ a disk-on-file approach, we have to run an off-line clean-up script that manually clobbers files for any deleted rows.

Even still, is there a way to bind a paramater to a filehandle, and have the driver figure out the details to get that file into the DB ?