"Base64 (a 4-for-3 encoding) wastes 33%, so I thought about perhaps
inventing Base128 (8-for-7) or maybe even a higher level to cut down
wasteage, but Base64 has the advantage that it avoids UTF8/ISO-8859
confusion and I bet Mr. LZW will eat that 33% anyhow..."
The thing to consider here is that, although XML is Unicode, the typical
encoding of it is predominantly ANSI octets, in which case Base64 works
pretty well. But run some tests...