E. Hazeltine, D. Teague, and R. B. Ivry have presented data that have been interpreted as evidence against a central bottleneck. This article describes simulations of their Experiments 1 and 4 in the ACT-R cognitive architecture, which does possess a central bottleneck in production execution. The simulation model is capable of accounting for the emergence of near-perfect timesharing in Experiment 1 and the detailed data on the distribution of response times from Experiment 4. With practice, the central bottleneck in ACT-R will be reduced to a maximum of 50 ms (1 production cycle) and can often be much less, depending on timing of stages and variability in their times. The authors also show, with a mathematical analysis of E. Hazeltine et al.'s Experiment 2, that the expected dual costs for these kinds of highly practiced tasks will be small in many circumstances, often under 10 ms.