We are currently witnessing the emergence of two important trends in wireless networks, namely the increased usage of Internet like applications, many of them which are delay tolerant, and cognitive radio techniques. In this paper, the focus is on how to capitalize the delay tolerance of various applications (such as email, Peer to Peer networks, social networking & operating system updates and file transfers to mention just a few) to reduce the energy consumption in cognitive networks. Since it becomes feasible to estimate the Primary Users (PU) connections for the Secondary Users (SU) by contacting a trusted database containing the information of PU traffic, a scheme is proposed that explicitly utilize the distribution of SU traffic loads to provide load-balancing. By modeling the problem under investigation as an M/M/K/L queuing system, the performance of the SUs competition is analyzed under various traffic blocking thresholds and queuing delays. To this end, by optimizing the frequency channel utilization, the number of SU connection that can be accommodated simultaneously is derived. A wide set of numerical investigations reveal how message transmission delays and the use of available white spaces can reduce the energy consumption without affecting PU traffic.