Code golf answers should be scored in whatever the interpreter can digest. If there's an X bytes file that does what it's supposed to do, you can claim a score of X bytes.
UTF-8 doesn't even apply to all languages. There are plenty of them (both older practical and newer recreational languages) that do not even understand UTF-8/Unicode, so encoding the source code in UTF-8 for scoring purposes is rather arbitrary. Other languages force you to use UTF-16. Again, UTF-8 doesn't make sense in this case.
Finally, there are languages that do not even have a concept of character (TI-Basic, byte code, etc.), and thus cannot be represented by any character encoding.