iirc its because core usage is an average over time, and you can get weird situations where switching between cores (which happens frequently depending on the scheduler) causes the average usage of one core by the program to not be zero yet, and the usage on another core to already be 100.

This is purely speculation though