Current optimization logic (https://github.com/evernote/serge/blob/4fde09810df74305f3ba33f94986cb209f1ba05d/lib/Serge/Engine.pm#L1202-L1206) doesn't account for the fact that the TS files for language may need to be rebuilt if there are new potentially reusable strings in the database for that language.
Right now strings will be reused only when there's another event that will trigger the update: source or TS file is changed, or were just created, or target file is removed.
When translations reuse feature is enabled on a job, an extra check needs to be done (for example, a highest usn of the language needs to be compared, or file cache needs to be analyzed).