As far as I can see the question being addressed is which models best explain experimental results (at equivalent model complexity). If quantum(-like) probabilities pay off let's use them.

Up to now the models have mostly been a heap of heuristics but those produce models with lots of degrees of freedom that don't generalize well.