When multiple tasks are ongoing, the ideal speed is not necessarily the sum of all the speeds for all those tasks. This is because power is not a linear function of speed, so superimposing schedules consumes a different amount of energy than running them sequentially. Unfortunately, computing a reasonable speed schedule that is the conjunction of two is extremely complex, so we avoid the issue by simply running at the maximum speed available when there are multiple tasks, and continue at that speed until no tasks remain.
Fortunately, in a mobile computer (and frequently in a desktop computer) there is only a single user and typically he will only notice the performance of the task with which he is currently actively involved. Therefore, typically there will be only one ongoing task at a time. Evidence supporting this comes from workload analyses we performed in [12] on months-long traces of eight desktop computers. We found that, depending on the user, between 94.7 and 99.3% of all user interface tasks finished before the next one began.