We characterize the asymptotic behavior of the throughput capacity and packet delay in cognitive radio (CR) networks. For the model where the network density grows, we derive the scaling laws for secondary users (SUs) who opportunistically access the spectrum when primary users (PUs) are idle. More specifically, we assume that SUs do not know PU (transmitter or receiver) locations a priori, and access local channels only after sensing them idle. We consider general overlay networks with arbitrary scaling of the number of PUs and SUs, and assume that PU transmissions are on and off intermittently in a dynamic network operation. Accordingly, the asymptotic SU performance is characterized in terms of the average area covered by PU transmissions. It is clear that shortest path routing (SPR) yields higher capacity whereas multipath routing (MPR) yields smaller delay. When compared with SPR and MPR, our findings show that source-based network coding (NC) with opportunistic routing, achieves the same capacity performance as SPR and the same delay performance as MPR.