Sometimes, the profiler seems to pick a random component to account for the missing execution time of the definition.
Except this time, my component has an alibi : it couldn’t run because the trigger was always “false”.
This trigger is activated through Human UI, and I know I didn’t press that button.
So it seems to take the total effective execution time, substract to that the sum of all the execution times of the components and pick a scapegoat to be responsible for the difference.
By default, Pancake’s perf analyzer just displays timings from GH’s profiler, similar to MetaHopper’s Bottleneck Navigator. So theoretically in this case it won’t interfere with GH’s execution pipeline.
As for your issue, Grasshopper’s C# component needs to call the old CodeDom compiler, which can be slow under certain circumstances, e.g. when the host NGEN is busy.