It was difficult to even recognize as a bug. Apparently the problem is rooted in a miscommunication between a higher level abstract part of the scheduler which recommends sets of "ideal cores", and ...