In private followup discussion to this thread it occured to me that I 
implicitly understand UTF8 to mean a UTF8-based variable length encoding
for up to 4 byte characters (what we have as wchar_t now) - not the
popular version that encodes Unicode characters only.

I don't know anything about character histograms in texts using non unicode
characters, so maybe this encoding will get unwidely long. But I don't know
any other (popular) variable length ASCII-clean encoding, and I think we can
not change all the char* in the traditional APIs.

Plan9 had some papers with discussions about this - but they restricted
themselfs to Unicode.

Martin