vtkCamera ModelMatrix is not fully supported
A number of classes (such as pickers) use the focal point, positions etc of the camera to drive their algorithms. When you set a model matrix on the camera these values (FP, Pos) are no longer in the right coordinate system causing picking to fail. Or put another way, the pickers do not apply the Model matrix in the Camera to the props bounds to get their bounds into world coordinates.
The old model was
actor bounds -> (the actor's actor to world tform) --> (camera world to View)
with someone setting model matrix on camera it looks like
actor bounds -> (actor's actor to model tform) --> (camera's model to world tform) -> (camera's world to View transform)
and there is a lot of code that is built around the old model and does not handle a model transform on the camera. If you search you will find few/no tests that set a scaled/translated model transform in the camera.
I personally would yank the model transform on the camera as it does not make sense to me but either way it is currently broken.