This is a figure from a meta-analysis that Heather Leary and I finished earlier this month. It's about tutor variables that moderate problem-based learning outcomes. I should add these are cognitive outcomes only for students (standardized tests, essays, etc . . . ). Each of the outcomes compared a PBL treatment with a "control" or "lecture" condition (although those can mean many different things). There's a lot going on here, the little (n=xes) need some attention as they show the number of outcomes. The very rough scale for effect sizes is .2 = small, .5 = medium, and .8 = large.
Many of these combinations don't have what you would call overwhelming amounts of evidence, but we do know a lot about faculty whether they received formal training as PBL tutors or not, and the difference in terms of student learning are non-existent. We have no causal data, which I can't emphasize enough--so maybe this is because faculty are immune to training, falling back on their lecture roots, maybe the training was poor, maybe it's because of Turkey burgers--but it's pretty shocking. My vote is for faculty resistance, other research points to epistemological beliefs being pretty stable by the time you hit undergrad. And for many, this kind of approach requires a fundamental shift in epistemological beliefs. There are documented cases of faculty tutors going into a small group PBL sessions and just lecturing, even after training. But research is not a democracy. Somebody could make a serious name for themselves figuring out why this is the case.
