Cluster sampling loses to SRSWOR exactly when $(M-1)\rho>0$

A simple random sample of $n$ clusters is selected from a population of N clusters each of size M, then cluster sampling will be less efficient than SRSWOR if ($\rho_d=$ intra class correlation coefficient between elements belonging to same cluster):