Cool test, Kadri and thanks for the tip, very useful! I love that crash actually, and have to check that out in depth.
Afaik recalculating normals takes the angles of the polygons and makes a weighted new set of data. If you set the angle low, the overlap of angles is somehow not smooth, so I guess that's in the numbers somewhere. All I know is that you have to choose the angle just right for different parts.