The goal of the flexible, efficient and fair spectrum allocation is to increase the spectrum utilization of radio resources in future wireless systems. According to the cognitive radio (CR) concept, the nodes are expected to sense their radio environment, take decisions on their operation in the network and learn from their past actions to better adjust to the network dynamics process in various unplanned situations. The CR node can take actions resulting from the input information processing, although in most cases this information is incomplete or inaccurate. CR node may acquire the necessary information needed for its efficient operation by accessing the control or management channel(s) or by interaction with other nodes. Unfortunately, control channels may not always be available and the neighbouring nodes may not be interested in cooperation due to the cost of the spectrum and energy resources. Therefore, the major challenge for a CR node is to operate efficiently with incomplete or limited knowledge on the network and cooperate with its competitors. For CR, game theory provides interesting tools to study competition and cooperation among rational and intelligent players taking decisions with limited or incomplete information. CR nodes can exchange information, cooperate or learn because they were programmed to perform such tasks. Such processes have the associated cost, usually expressed in consumed energy, time or spectrum. These costs must be balanced with benefits. Therefore, the GT models must be carefully selected and evaluated for the application in resource sharing to comprise with practical limitations of the dynamic CR networks. In this work the practical issues of cooperation among cognitive radio nodes competing for available resources in the decentralized networks are considered. It is pondered how the theory of competition and cooperation (game theory) meet the practice, by discussing the quantitative metrics of the cost of avoiding cooperation (the Price of Anarchy-PoA), of having limited knowledge of the competitors (the Price of Ignorance-Pol). Some practical approaches to the spectrum sharing and allocation problem are also presented, which make use of representative, intentionally reduced information that the CR nodes have to exchange. One of the presented methods is based on the repeated game against the network-nodes community using the aggregated knowledge of its possible behavior. The other one is based on the coopetition methodology, which combines the advantages of both cooperative and competitive approaches. It is shown that the problem of radio resource allocation in wireless systems can be solved efficiently by using these not-optimal but practical approaches, by presenting some indicative results: the information-data sum-throughput, Jain's fairness index, PoA, Pol, and the network welfare function equal to the sum-throughput net.