The decentralized Cognitive Radio Medium Access Control (CR MAC) protocols proposed so far are usually evaluated considering homogeneous sensing conditions where all Secondary Users (SUs) share the same view of the channel. By doing so, realistic heterogeneous channel sensing conditions are neglected, which happens due to different spectrum sensing errors and/or different patterns of channel occupation by the different Primary Users (PUs). This work aims to study the performance of CR MAC protocols when different SUs competing for the medium sense the channel differently, i.e., when they achieve different channel sensing decisions at the same instant. The study includes several simulation results comparing the performance of two random access (RA) contention-based protocols and one reservation-based (RB) protocol in terms of throughput under homogeneous and heterogeneous channel sensing conditions. By adopting a systematic approach to characterize the heterogeneity of the SUs' sensing outcomes, it is shown that an increase in the level of heterogeneity decreases the aggregate throughput of all protocols. The results also indicate that under homogeneous channel sensing outputs the RB protocol outperforms the RA protocols. However, due to reservation impairments caused by the heterogeneity, it is shown that heterogeneity causes a higher throughput loss in the RB MAC approach.