I’ve moved this over to the Technical support forum as it fits there a bit better now (it’s either that or the Python support forum).
At any rate (and again, remember that my understanding of the MDL0 format is shakey at best… in fact, looking at your spec, it seems I was wrong about the format not supporting bone animation), I need to ask more questions. Specifically, does the object you’re attempting to import have any animation? More importantly, can a vertex have more than one transform state (that is, if the first image you post is the base and the second image is the vertices after transforming, could you have a third version with the vertices located at different coordinates)?
Now, assuming that you can, in fact, have more than one transform state for vertices (are they named?), then the translation to shape keys is pretty straight-forward. On import, you take the initial location of vertices and you set that as a basis shape. Then, for each transform state, you create a shape key and move each vertex accordingly. On export, it’s the same process, but in reverse.
However, if I’m completely wrong and there’s only one state of transform vertices, then shape keys are not your answer. In that case, your best option is probably to pull in the raw vertex locations first and then simply run a second pass where you move those vertices based on the values stipulated by the transform vertex values.