Nowadays it's less of an issue, but back in the days windows, mac and linux all used different default encodings. This meant that opening and saving the source file in one of the system could mess it all depending on the editor and how careful the devs are.
Also, there is the canonical writing issue, where you can write different utf-8 strings that all end up displayed the same. It's not an issue if it's just display text, it's a different story if you have a method that doesn't get called because the byte sequence was different, or more subtly hash indexes that seem to match but actually don't.
BTW there are companies that write some of the UT methods right in Japanese, as it matters less, and it's quite fun as there is less restrictions on what can be put in the identifier (full-width parenthesis, spaces, comma etc, are allowed in that context)
Just for the record, in Russia, we write all identifiers in English and practically all comments in (sometimes broken) English as well. Russian comments seem heavily discouraged, but hey, it's possible there're places where there're not.
I am not from Japan, but do work in two languages other than English (both official here). This is pretty typical actually. English names for identifiers (sometimes badly translated or simply transliterated), local languages for comments.
Don't most modern compilers allow utf-8 characters in symbol names?