With the recent additions to the Compact Language List it has become better, but it is still not optimal. It adds languages that is never used, which takes slots from more important languages.

One solution could be to make a tracking hash for languages in local storage, where we add "one" for each page language each time the user opens a page and 1/n for all n languages shown by the CLL. If the number for a language drops below a threshold then the language is removed and the slot become vacant.

This should only be done for languages the user hasn't explicitly told the system to include, for example by listing them in babel boxes. (Note that babel boxes isn't used as of this writing.)