@soenke.heuer - also remember that white balance may affect the highlights. I got ok results with -H 2 and a WB performed on patch 4 of the last row. This WB is performed in rawtoaces using:
--wb-method 3 <x y w h>
where x and y are the coordinates of the upper left corner of the WB box and w and h are the width and height of the box. You can specify a box that includes pixels in a neutral patch to base your WB for rawtoaces libraw conversion.
I found that this method was ok, but produced WB that appeared and measured too blue. Give it a shot, see if it works any better with -H 2 in producing neutral highlights.
As Thomas already pointed out, at some point there is too little data in the blown channels to produce a reliable highlight reconstruction.