Thx again, but it doesn't seem to be a problem with the poly count - the char models are just buried in a "mesh coffin". And it's the same with vehicles:
I tried an exclusion depending on the vector length of vertices but guess I'll have to wait for another brainwave.
Though I could create an UV extractor. For simple models the UVs seem to correct; for others (see in above pic) setting of some points to 0.0 0.0 might be required but that would be easier than recreating the complete UV map.
edit: here I deleted all submesh_1 faces with vertices greater than ~2900
UV map Jackie (maybe uncomplete):
(If ripping the models and fixing the missing parts were acceptable we would be done here.)
edit: well, combining the above mentioned face data with the ripped model data would need some recalculation, I guess. This being a task I haven't ever done so far.
A simpler solution would be to merge the ripped model into the extracted one in a 3D program. Then move/rotate it until it's fitting to the buried one and deleting this one (only vertices, preserving faces). Didn't try such thing so far, too. But should work. More or less...
edit2: I tried to exclude vertices with the same vector gradient. At no avail.
And then I realized: the parallel vectors I tried to eliminate just appear because of an position offset between the model "quarters". I'm such a dumbo...

Maybe the vectors building a quarter are grouped in the data? After I will have found them adding an offset would be a lesser task...