In relation to T177871 (and before that T147959), we discussed reviewing the language analysis config for "other" zh-* languages. None of them were affected by the changes in fallback language usage, and it's a non-trivial question, so I'm creating a new task to be prioritized separately.
"Chinese" refers to a family of languages, not all of which are mutually intelligible. The wikis in these languages (and the writing systems they use) are listed in a table on English Wikipedia. (Note that as of the time of this writing, Classical Chinese is discussed beneath the table.)
The relevant languages don't all have codes that start with zh-, so we should look at zh-min-nan, zh-yue, and zh-classical, but also cdo, wuu, hak, and gan.
Note that several of the wikis use Latin romanization in addition to either Traditional or Simplified characters.
A quick survey of a couple dozen random articles on zh-min-nan.wikipedia.org show that all are written in romanized Chinese.
A quick test on (Chinese) text from today's zh-yue.wikipedia.org front page show that the CJK analyzer generates overlapping bigrams, while the ICU tokenizer generates tokens in a variety of lengths (1 being the most common, 14 being the longest).
So it's likely that switching to the CJK analyzer would be better, but I'm not sure if we should do anything for the ones that use both Traditional and Simplified, because I'm also not sure if the S2T conversion we use for zhwiki will do the right thing for other varieties of Chinese.
I also happened to notice that Zhuang/Vahcuengh (za) is a Tai language mostly written in Latin characters, but with a lot of Chinese characters. We should look at that, too.
Anyway, it's complicated, so it needs its own ticket!