I would think that the migration overhead would depend more on the interconnection network and less on whether the multicore is homogeneous or heterogeneous.
It depends, on hardware supported Shared Memory (i.e., multi-core single node) very little, however in Distributed Shared Memory (DSM) and Distributed Memory it is by far much higher due to interconnection network and other costs like hidden costs like remote memory accesses in DSM.