Power saving and management is a critical aspect of cooperative spectrum sensing in cognitive radio networks, in which a group of secondary cognitive radio nodes share the task of spectrum sensing in order to detect gaps in the spectral bands temporally or spatially not utilized by primary transceivers, facilitating dynamic allocation of physical layer and networking resources that finds application to emergency networks and the military, among others. We show how the problem of power-optimized cooperative spectrum sensing can be cast to a monotonic class of restless bandit processes and present analysis and computationally efficient solutions. Our numerical results verify the effectiveness of the proposed power-optimized spectrum sensing scheme.