If you assign preferred execution attributes to a critical user and maintain default attributes for other users, Adaptive Server does what it can to execute all tasks associated with the high-priority user first.
One result of scheduling is that, when an engine does not find a task in its local run or a global run queue, it attempts to steal a task from another engine’s local run queue. Engines can steal tasks only with a normal priority, and can never steal a high-priority task for high-priority users. If the engine’s loads are not well-balanced, and the engines running high-priority tasks are heavily loaded, the act of stealing tasks can lead to starving high-priority tasks of CPU, which is opposite of the intended affect of scheduling, but a natural side effect.