What happens to cross-modal perceptual resources when task performance is predictably poor? This question is central for understanding the efficiency of attention allocation among concurrent tasks in different modalities and for assigning multi-tasking operations in real-life situations (such as texting while driving a car). General load theories propose that attention allocation may be a zero-sum game so that the more difficult one task, the poorer performance on a second. Consistent with this view, having detected a first target, observers often fail to report a closely following second, the Attentional-Blink. What happens when different senses are involved? We tested the novel prediction that attention can be shifted from tasks that will probably fail anyway, to improve performance of a concurrent, more promising task. To test this, participants performed auditory pitch discrimination while engaged in detecting two successive visual targets in an Attentional-Blink paradigm. We found that, rather than suffering from general system load, auditory performance was significantly superior during the Attentional-Blink. We conclude that when visual task failure was predicted, resources were shifted to the auditory system. These findings suggest presence of a highly sophisticated attention–allocation mechanism that computes likelihood-of-success versus cost for each task on a moment-to-moment basis.