Generally, when you're doing string manipulations inside XSLT/XPath,
there really is no such thing as ISO-8859-1, UTF-8 or any other
encoding, since the "string" data type in XPath is just a string of
Unicode characters.  The encoding of the input is used to map the
sequence of octets to Unicode characters on the way in and the
requested encoding of the output is used to do the reverse on the way
out.

Percent-escaping is sort of an exception since it is, really, a form
of encoding, but not one that is generally handled automatically by
the parser, serializer, etc.  So, you may need to decode the
percent-escapes, but you shouldn't have to worry about the overall
encoding.

If you think your use case requires that you really do need to deal
with encodings, please tell us a little more about it, so that we
might be able to better suggest a solution.  How is this string
getting into your transform while still being encoded?
...