The bytes are completely trivial, unless your entire document is a non-latin script.
And my point is that the UTF-8 rule is not always followed, because source code gets passed around via e-mail, copied from browsers, edited in a console, saved with Notepad, and so on. Ideally file content would only ever be touched by source code editors configured only to ever operate in UTF-8 mode, but in the real world with real development teams and new interns, that is an impossible thing to ask.
At my last job, I made sure all accented characters in source code were only ever written with entity references, and we went from having weekly problems with accents somewhere on the site, to almost never.
If that describes your development process, you have problems that entities won't correct. I think you maybe misunderstand the audience here: these are guidelines for Google's engineers. Honestly I think demanding (and assuming) that your developers use a toolchain that doesn't suck is a reasonable choice in that environment. It may not be in all shops.
