some more details (before I make a test case)
the glyphs of the real font have an expansion_factor of -10000 (post_linebreak)
and the virtual fonts have an expansion_factor of 0 (so the question is: why).
The PDF also shows the expected differences:
/F11 9.99999 Tf 0.99 0 0 1 28.346 804.543 Tm
/F13 9.99999 Tf 1 0 0 1 28.346 804.543 Tm
(see the difference in the first argument to the Tm and the different kerning)
So I will investigate further where I messed up.
Thanks so far!