Okay, I have the same problem on compressor. PTC explain me that is due to the "hyperstatisme" of the model, but I never find the solution.
There is an "easy" fix for this situation. I have modeled many engines and complete valve trains in the past and the one key element to obtaining accurate bearing support loads is redundancies. You must create a set of joints that will not generate any redundancies. You can measure "system" quantities like DOF and redundancies as a measure to make sure you're OK. The trick is creating 6DOF joints for all your bearing supports and applying a heavy spring or servo motor to constrain the radial directions. When you do this, redundant supports can be calculated accurately. I say it's "easy" because the process is simple, but it is more involved than applying pin joints everywhere. You'll need to make a co-ord system on the shaft and bearing/housing in the same locations of the center of the bearing (so their centers and rotation axes line-up exactly upon assembly). You'll also need datum points on both parts at the center of these co-ords to reference for translation "zeros". Apply servo motors to the X & Y (radial) directions and make sure to add these to the dynamic analysis definition. You will obtain accurate bearing loads. I did this many years ago while developing an engine bearing system for a very large European automaker. They ended-up re-formulating their calculations 3 times until theirs matched mine ... they agreed errors were found in their calcs - twice.
BTW - the new "Bushing Load" in Creo 2 may be a good substitute for my 6DOF approach - I just tried it for the first time a couple days ago. I didn't seem to get the same results as using 6DOFS, but I need to spend more time with it before passing judgment.