Hi Peter and thanks again for the answers.
Maybe I had a poor choice of words: when saying dynamic, I was thinking about kangaroo: the deformed mesh had to undergo a dynamic relaxation, for a more physically-correct representation. But as I analysed meshes more and more, it appeared that the errors I encountered were due to the mesh being a tiny bit wrecked (non manifold edges and 0-area faces). Once this was fixed (with a more delicate approach with code), the problem went away 
The other problem I am encountering is not exactly the technical “how” of moving the faces, I had already managed to write down some code doing so (though more error-prone than yours, I’ll admit 
). The face classification I had made as well.
My real problem would be more (refering to the 2nd post of this night): “how to choose which vertex of the mesh to move, and where to move it to in order to reach the goal, without wrecking the mesh ?”
(Fit deformed mesh to its original Brep shape)
I have been trying considerations such as looking the faces that are not contained in one of the surfaces of the base shape brep and keeping an update list of that throughout the script, then considering faces where only one vertex had to be moved (ie already 2 vertices on a same surface and the 3rd one on another surface), etc. not very successful.
At least not totally: for a good number of faces fixed, I get a few number of completely broken (and without any means of fixing them as the tolology of the mesh looses itself when it gets broken).
But a solution you seem to offer would be to do it manually (since my already attained result is humanly workable) and move the vertices one by one, taking care myself of moving them in order not to wreck the mesh. The “getting rid of the static aspect of GH”.
However, for this part I have not understood how it could work, with the example scripts you posted 
Would you have any further explanations, or suggested paths ?
Many thanks !
EDIT: I will just go and edit the mesh manually in Rhino with a little PointsOn actually, much simple and could have saved me a couple of days work …
Still interested in the interactive aspect however!