After some experimentation and some RTFM-ing I realized that of course I wanted 'bytea' fields instead of 'text' fields.

Currently we use a Documents table with a bunch of information in it as part of our system; including the file itself would smooth a lot of things out -- we'd have to copy the file from Postgres to the file system for processing but (this is the cool part that I love) any files that were created during processing can then be stored back into the database. This means that we have the ability to back-track to previous iterations. (For obvious reasons, I can't describe my system in a lot of detail.)

Your advice (third paragraph) does make sense .. thanks for the pointers. I'll be reading up on SPOPS more this week to see if we can fit your wheel in somewhere rather than build our own.