I mentioned in another reply that it actually does gzip fairly well (and bzips even better). But you still end up with an order of magnitude expansion with gzip (as compared to gzipping the original source), and around 3x expansion with bz2 (again compared to bz2 on the original source).

That's from a <0.5KB test input, so the expansion might be mitigated a little more for larger files. I was going to test on a 3KB microlibrary, but gave up after about 10 minutes of waiting for the conversion to finish.