I don't think this question can be answered in general, because ultimately eye-tracking tests only capture the movement of the eyes of the user and cannot tell you exactly why the user is doing something (as this is the same with all user testing).
The way eye-tracking tests have often been used, as you pointed out, is to provide support for certain decisions to be made, and often the user tests are 'rigged' to provide the desired result. That is, the experimenters often intentionally or accidentally introduce biases into the testing and analysis of the results in search of an answer that they are expecting.
So we already know from existing literature in the field of psychology how eye-tracking has been used to pick up signature movements when doing analysis of people's expression, including eye movement to detect signs of deception, and it has also been used to pick up scanning patterns when analyzing people's perception of their body image in mental illnesses like anorexia nervosa.
It seems like these applications are effective because they are trying to pick up the subconscious movement of the eyes and trying to separate them from the conscious movements, which can be influenced by a number of factors. So if you can create tests that rule out a large number of variable factors then it is more likely to be effective for user testing of the hypothesis that you are testing.