I donwloaded the MRB and the associated rigid transformation matrix has a huge translational component that seems unrealistic in relation to the scale of your data:
If you apply that transformation to a target model, it moves out of the scene. This might be one of the culprits.
Can you share the original models that you used to derive the point cloud? We need the models to test it for ourselves and see if we can replicate the issue.
@smrolfe @agporto