Several cognitive processes are involved in task-switching. Using a prosaccade/antisaccade paradigm, we manipulated both the interval available for preparation between the cue and the target and the predictability of trial sequences, to isolate the contributions of foreknowledge, an active switching (reconfiguration) process, and passive inhibitory effects persisting from the prior trial. We tested 15 subjects with both a random and a regularly alternating trial sequence. Half of the trials had a short cue-target interval of 200 ms, and half a longer cue-target interval of 2,000 ms. When there was only a short preparatory interval, switching increased the latencies for both prosaccades and antisaccades. With a long preparatory interval, switching was associated with a smaller latency increase for prosaccades and, importantly, a paradoxical reduction in latency for antisaccades. Foreknowledge of a predictable sequence did not allow subjects to reduce switch costs in the manner that a long preparatory cue-target interval did. In the trials with short preparatory intervals, the effects on latency attributable to active reconfiguration processes were similar for prosaccades and antisaccades. We propose a model in which the passive inhibitory effects that persist from the prior saccadic trial are due not to task-set inertia, in which one task-set inhibits the opposite task-set, but to inhibition of the saccadic response-system by the antisaccade task, to account for the paradoxical set-switch benefit for antisaccades at long cue-target intervals. Our findings regarding foreknowledge show that previous studies used to support task-set inertia may have conflated the effects of both active reconfiguration and passive inhibitory processes on latency. While our model of response-system plasticity can explain a number of effects of dominance asymmetry in switching, other models fail to account for the paradoxical set-switch benefit for antisaccades.