This paper presents a run-time slack distribution strategy for precedence constrained task graphs mapped on to a voltage scalable multiprocessor platform. Online dynamic voltage scaling techniques use the slack formed due to early finish of tasks and lower the supply voltage suitably so that energy reduction is achieved while meeting the worst-case deadline of the task graph. Given a task graph mapping on a multiprocessor, the proposed offline analysis phase calculates expected slack and expected computation ahead of each task. The online voltage scheduler uses these values and the current slack to decide the voltage and frequency of execution of the next scheduled task in order to minimize expected energy dissipation. Results show improvement in average energy saving with this methodology over most of the presently known online techniques