I am using a UTF-8 locale and all coreutils except for tr seem to agree
on that. However, tr behaves differently and seems to always assume that
1 byte == 1 character even in a UTF-8 locale. Consider this:
IMO, this clearly indicates that tr considers each character to be
exactly one byte wide. In the case of "tr äöü aou" all three umlauts
have the same first byte, so the last substitution for that byte is
effective, that's why ä becomes uo and not ao in that case.