The difficulty in combining weighting and selection formulas based on different models is that the intermediate variables necessary to calculate something in the vector model are completely unusable for calculating something in the probabilistic model. Thus, the real question becomes whether or not the performance increase outweighs the time and space tradeoff for the additional overhead.
In Harman's 1992 followup study, she looks at some probabilistic term selection formulas, in particular the wpq formula, using two different variations for vector-based term weighting. Her findings indicate that both variations of the wpq formula perform without significant difference from the other three top selection formulas studied. Efthimiadis' findings, summarized in Section 5.3.2, are similar, although he suggests that his r-lohi formula works well all around.
Many studies show that trimming the number of terms used for expansion results in an increase in performance, although no study has conclusively found a single selection formula that works well in all circumstances. In most cases, they find two to four formulas that tend to work equally well. With that in mind, it is likely that combining weighting and selection functions from different models will not result in any significant performance increase over using selection and weighting functions based on the same model, and that the added overhead will make the combination unattractive to operational systems.