In this article, the parallel multiple access channel is studied under the assumption that transmitters maximize their individual spectral efficiency by selfishly tuning their power allocation policy. Two particular scenarios are studied: (a) transmitters are allowed to use all the available channels; and (b) transmitters are constrained to use a single channel. Both scenarios are modeled by one-shot games and the corresponding sets of Nash equilibria (NE) are fully characterized under the assumption that the receiver treads the multiple access interference as noise. In both cases, the set of NE is non-empty. In the case in which transmitters use a single channel, an upper bound of the cardinality of the NE set is provided in terms of the number of transmitters and number of channels. In particular, it is shown that in fully loaded networks, the sum spectral efficiency at the NE in scenario (a) is at most equal to the sum spectral efficiency at the NE in scenario (b). A formal proof of this observation, known in general as a Braess paradox, is provided in the case of two transmitters and two channels. In general scenarios, we conjecture that the same effect holds as long as the network is kept fully loaded, as shown by numerical examples. Moreover, the price of anarchy and the price of stability in both games are also studied. Interestingly, under certain conditions on the channel gains, Pareto optimality can be achieved at some NE if and only if the number of channels equals or exceeds the number of transmitters. Finally, simulations are presented to verify the theoretical results.