Your extra "confirm" makes me retype this message hours later, because it expired. Anyway, when saving as UTF-8, 7 or 11 bits per character give the optimal sizes, 8 & 12 bits are the worst, due to the design of UTF-8, but not quite a 10% savings at best. It was easy to change the python port to store different numbers of bits per character, so I did that for testing. for UTF-16 storage, 15 bits works well, giving a 20% savings versus the uncompressed version in UTF-8. So I wind up with different application specific encoding techniques that have saved me about 25% instead of using your scheme. But I like your scheme, and may use it if I need to do significant local storage. At the moment, a few hundred bytes is all. If you want my tweaked code, send me an email address to send it to.