Unfortunately, I knew the issues of distributed solver and tried to run the simulation with only one core aka undistributed solver. The issue is not caused by solver distribution.
I guess the reason is the element status at the beginning of the first substep of the second step. During the first, heating step, the workpiece elements are all deactivated (via ekill at the beginning of the first load step). In the second load step, the former deactivated elements are activated via ealive.
Unfortunately, the simulation is not running anymore without the /post1 trip in every substep (on Ansys WB R2021 R2 Teaching) a few weeks before the model without /post1 script was running just fine.
Other simulations are running without error.
I don't understand whats happening here, seems the file is corrupted.