How reliable are empirical genomic scans for selective sweeps?