I feel a certain responsibility to help with this because I think I came up with the original rig a while back, which has since been made more complex and had other elements and parts added to it. It's certainly very complex now, and not helped, as you note, by a) viewport performance (xp calculations mainly) and b) looping an area not at the start, meaning the program takes a frame or 2 to catch up with where it is meant to be in the animation. And all this makes diagnosing the issue / suggesting a workaround / alternative method most challenging. I haven't found an answer yet.
I would agree with previous posts recommending faking it if I thought that was going to be possible, which I don't. This is not a fixed camera as I understand it, and there is simply nowhere to hide any transition points in any of the objects concerned that won't be seen from some angle or other.
And whilst attaching nulls to the holes in the Governer Arm seems initially promising, the question remains as to how those nulls would be also attached to the various parts of connected models in a way that could move a group of points at the end of each spring or armature without affecting the natural look of the overall deformation imparted by the keyframed FFD controlling the spring length and shape, which is very specific, curves with the central arm, and must maintain that during animation as it stretches and contracts. Very very difficult.
CBR