It's possible to be smarter than that. I believe Chrome doesn't drop down to punycode if it sees the string is entirely within a single language so that there is no possibility of look-alike characters. It might even be checking explicitly for the presence of look-alike character pairs, since they're known ahead of time.
Just look at 田中太郎. Four completely independent glyphs, equal widths, no modifiers, in left-to-right order. It is a very good way to write things. The only issue is the fourth one seem to barely have enough pixels.
Yes! Finally a protocol that does not discriminate against non-ASCII languages (like punycode does in my opinion).
Hello, my name is xn--jeprenj-rqb! </rant> (: