In this paper, we study and analyze fundamental throughput and delay tradeoffs in cooperative multiple access for cognitive radio systems. We focus on the class of randomized cooperative policies, whereby the secondary user (SU) serves either the queue of its own data or the queue of the primary user (PU) relayed data with certain service probabilities. Moreover, admission control is introduced at the relay queue, whereby a PU's packet is admitted to the relay queue with an admission probability. The proposed policy introduces a fundamental tradeoff between the delays of the PU and SU. Consequently, it opens room for trading the PU delay for enhanced SU delay and vice versa. Thus, the system could be tuned according to the demands of the intended application. Towards this objective, stability conditions for the queues involved in the system are derived. Furthermore, a moment generating function approach is employed to derive closed-form expressions for the average delay encountered by the packets of both users. The effect of varying the service and admission probabilities on the system's throughput and delay is thoroughly investigated. Results show that cooperation expands the stable throughput region. Moreover, numerical simulation results assert the extreme accuracy of the analytically derived delay expressions. In addition, we provide a criterion for the SU based on which it decides whether cooperation is beneficial to the PU or not. Furthermore, we show the impact of controlling the flow of data at the relay queue using the admission probability.