Task switching requires to flexibly engage in the processing of one out of two tasks sets in each trial. Curiously, if one of the tasks becomes irrelevant (fade-out), performance in the remaining task is worse than when performed as single task. This so-called fade-out cost demonstrates that the suddenly irrelevant task cannot be discarded directly (Mayr & Liebscher, 2001). In the present study, we used list-wide proportion manipulations to induce different control states during task switching with a subsequent fade-out: first, at the level representing the task structure by manipulated proportions of task switches (Experiment 1) and second, at the level determining how well tasks are shielded by varied congruency proportions (Experiment 2). Results showed larger fade-out costs for groups of high switch frequency (Experiment 1), whereas the congruency proportion did not affect fade-out costs (Experiment 2).