>>10290
Hexadecimal, or uuencoding so each letter is 8/26 bytes rather than the more common 26/8 bytes where converting to text normally causes the binary to expand by about a third when posting to usenet.

Also, of course, it's not meant to be typed back in by hand; once you have a decent OCR-capable alphabet, you can double up on capacity by:
* use smaller font. size 6, if reproduced by good laser printer, is scannable back into data.
* use more characters. Since this isn't meant to be read by humans, there's no need to use the alphabet as such. Tiny QR codes maybe? Cyrillic characters where there wouldn't be any difficulty distinguishing even at size-6 fonts? One, two, and three-dot umlauts or other diacritics? Again this doesn't have to make sense to humans so all 42 letters of your made-up alphabet can now be multiplied by three using made-up diacritics, resulting in each size-6 character representing 21 bytes of reproducible data.

12,192 characters per page needed to shove a quarter megabyte per side of each sheet of paper.