Sure, UTF-8 sucks for Chinese and UTF-16 is bad at English, but in practice, high-codepoint languages are rarely mixed with low-codepoint ones. Notice that when sending an email many mail programs will select the most concise encoding that happens to encompass every character in your message and usually not UTF-8 or UTF-16.

Adding some Chinese (from wikipedia) does actually show UTF-n as worst-case compared to UTF-16 and UTF-8, at least. I got UTF-16: 0%, UTF-8: 5%, and UTF-n: 7%.

It would be good to have demo text that plays to UTF-n's advantage, so I don't have to copy-paste from someplace like jp.wikipedia.org myself :)

It looks like it preserves some of UTF-8's stream synchronization properties, but does it have UTF-8's wonderful property of being recognizable by simple heuristics to great confidence even for tiny sequences?

For Russian, the difference between UTF-8 and UTF-n is statistically insignificant.