With the settings from my previous post, I re-ran all of the different AB-slants:
This is already much better! Mean error about 6 %, Standard deviation about 14 %.
The next step would be to do another set of runs with more iterations and an imroved mesh transition between BL of the AB and the surrounding mesh:
Unfortunately, my free coure hours are all used up 
. I reached out to support for my options.
Thank you all for you help so far,
Jan