A function in AUCTeX for doing indentation looks at whitespace syntax
    for finding the first non-whitespace character (and so does
    `back-to-indentation' in CVS Emacs).  That means it will skip the "Ü"
    and delete everything from the beginning of the line to and including
    the "Ü".

    I removed this code in CVS AUCTeX which now only uses
    `back-to-indentation'.  In Emacs 21.3 this function does not look at
    character syntax but simply skips spaces and tab characters at the
    beginning of a line.

If AucTeX specifically wants to skip just space and tab, it could do
that explicitly, rather than calling back-to-indentation.  That would
work in the latest Emacs.

We could look at trying to fix this by giving different syntax
values to the unibyte non-ASCII codes.  The problem is that
(I think) different non-ASCII coding systems would want different
syntax values.  However, making codes 200-377 all "word constituent"
might be better than making them all "whitespace".