There were 3 stages as before and the levels reduced from [1 2] in the first stage to [0 1] in the second and third. Judging by the results, the first stage makes great improvements, whereas animations of the second and third stages show just tiny vibrations around the skull.
The conclusion is that the far more significant scaling stage is the first and the latter are negligible. I must point out that it will be interesting to try treating scale and translation not as independent. It will be interesting alternating between these two stages, e.g. performing translation, scaling, then translation again. At present, the first stage of translation does the best it can given a known discrepancy in scale.