Performing two actions at the same time usually results in performance costs. However, recent studies have also reported dual-action benefits: performing only one of two possible actions may necessitate the inhibition of the initially activated, but unwarranted second action, leading to single-action costs. Presumably, two preconditions determine the occurrence and strength of such inhibition-based dual-action benefits: (a) response set reductivity and (b) action prepotency. A nonreductive response set (given when all possible responses have to be kept in working memory) creates inhibitory action control demands in single-, but not in dual-action trials, and the ensuing inhibitory costs are proportional to the level of action prepotency (i.e., an action that is easy to initiate is hard to inhibit). Here, we set out to test this hypothesis by varying representational characteristics in working memory (namely response set reductivity and action prepotency) across four experiments. In Experiments 1 to 3, we compared (a) a randomized mode of trial presentation to (b) intermixed, but predictable fixed sequences of trial types and (c) a completely blocked mode of presentation. As expected, dual-action benefits were strongly present in Experiment 1, significantly reduced in Experiment 2, and absent in Experiment 3. This pattern of results matches our predictions derived from the assumption that differential inhibitory costs in single-action trials are the root cause of dual-action benefits. Crucially, however, the results of Experiment 4 (in which response conditions were only partially blocked) pointed to a secondary source of dual-action benefits that was inseparable from inhibition-based effects in previous experimental designs: semantic redundancy gains.