Just imagine that mitochondria are an API to a power system and the nucleus is the control system. Sometimes it works out that a foreign mitochondrial component (i.e. 3rd party) works well, or maybe there are even whole classes that work well. But we just don't know. Maybe they are directly swappable, but since we don't even understand all the processes that go on and what communication (exactly) takes place we can't say for sure, we just have to go on primitive black-box cohort studies.