For me, UTF-8 or UTF-16 is certainly an improvement over JSON's UTF-8 only.
JSON is actually UTF-8/UTF-16/UTF-32.
I much prefer requiring only UTF-8. UTF-8 is enormously more common and a single encoding is a significant simplification; the gains from UTF-16 are much reduced by the fact that markup almost always uses ASCII characters. If an implementation wants to support UTF-16, it should be free to do so, but as a requirement it is on the wrong side of the 80/20 line by a long way.