Related: Fair size comparison across languages with different source alphabets
Scoring non-ASCII languages in code-golf has been long-debated. Since I primarily uses APL for code golf, I have seen and been in these discussions.
I think it is reasonable to handicap non-ASCII languages based on expressing power of a single character. However, utf-8 does not do that. Take APL as an example. All non-ASCII character APL uses are in the Miscellaneous Technical block which is 3 bytes. There are also ASCII characters that simply goes unused or for things that are useless for code golf (like forking or object-oriented stuff). Actually, APL uses only 50 or so non-ASCII symbols so APL programs can be mapped entirely within ASCII if you restrict legal identifiers and/or allow non-printable ASCII.
Community consensus is that any language whose spec is up before the challenge is posted is allowed if the OP does not otherwise restrict its use. And that leads to my question: Does a language which is only character substitutions of an existing language allowed? And, if it is, is a translation table for such a language sufficient as a spec?