Parallel switch on IncidentRadiation has negative effect on Win10-64bit-32core threadripper

parallel is on 4.6m parallel is off 3.4m

Weirdly the parallel switch makes the calculation substantially slower when On than Off…
When I run the same simulation on my old imac it runs super fast…in a few second, so something is clearly of…but what?
It’s just a simple shadow calculation using the latest LBT 1.1.0
Can anyone tell me what I’m doing wrong here?

thanks
anders