I think it is more than just the magnets.

I have written a new toolset just to debug the effect on morphs in the world coordinate space. I found that actually the layout of the whole virtual space is changed in a non linear fashion. Both the dialing itself as well as the axes of the (hyper)space appear to be unstable (from a linear point of view).  This is a fundamental issue and might be not the fault of poor v4. I hunted the issues down to m4 as well.

Poser seems to calculate in a way that I am not able to find out. It might be sums of rounding errors or be the cause of the core of their algorithms.

However I have a small hope. By repeatedly applying my algorithm and tracking the movement of the vertices in poser I seem to be able to reduce the error from about 10^-2 down to about 10^-6.  At least for the worst example it worked.

I morphed M4 posed with AdamTwaithes Pose 72 into a V4 with Adam's Pose 84 and the error was estimated at only 10^-7.

If this would work in general than it would get as close to mathematical perfection as possible :).
However I have to redesign the script from scratch now, and noone can say if it works good on any other poser than mine. @Testers: You will have mail soon ;)