What causes residual dual-task interference after practice?

Practice can dramatically reduce dual-task interference, but typically does not eliminate interference entirely. Residual interference after practice is especially large with certain non-preferred modality pairings (e.g., auditory-manual and visual-vocal). Does this residual interference imply the existence of a persistent central-processing bottleneck? To address this question, we transferred participants with previous dual-task practice to a psychological refractory period design. Although we observed residual dual-task costs in all four experiments, there was no evidence for a bottleneck, even with non-preferred modality pairings. We conclude that practice can eliminate the bottleneck limitation, but performance is still subject to other sources of interference, such as competition between central codes of the two tasks.