When you need to re-segment (paragraph-based > sentence-based) a TMX file and you've noticed that (some) source sentences correspond with more than two target sentences (or the other way around), you'll cannot do this re-segmenting on the fly (that is: during your current translation project).
Instead, you'll have to open the TMX file in TMX Edit Mode. Execute a Find and Replace action to mark sentence endings where necessary (see here).
Make sure that you use a unique number combination to mark the sentence endings, e.g. 000.
Run a QA for numbers. Remove the unique numbers that you've used to mark the sentence endings.
Open the TMX in Edit Mode:
Tag the sentence endings:
QA for numbers:
Keep the filtered result, remove the number tags via Find and Replace.