I have not gotten a chance to test your theory on multithreading, but wanted to go ahead and provide an example project for you. I've exported a project with 2 variants in the project folder: VC0 was created first and has the bug in it; I then created a copy of the variant, and did what I described above and the near shading loss/electrical shading loss increased by a combined 0.55%, as shown in VC1. It's not letting me upload a zip folder on this thread though, is there another way I can send it to you?
Also, I thought it was worth noting that my suspicion is that it may have something to do with the .csv ground data in the shade scene because I was not able to recreate the bug without adding the surface in.