Comment by andjd

I would be curious how this compares to a more-or-less off-the-shelf text compression algorithm like gzip. My guess is that over the entire database, this would be more efficient than the OP's ad-hoc implementation or any alternative mentioned here.