Typical job scheduling for parallel machines is done on a one-by-one basis because of the dynamic arrival of jobs and the different priorities of the jobs. In the general case, this does not provide the optimum solution from a global point of view. Thus, we propose an approach which selects a group of jobs from the waiting queue and hierarchically optimizes a longer-term scheduling plan. This becomes feasible because the approach is embedded in our Scojo-PECT coarse-grain time sharing scheduler which separates jobs by priority. We demonstrate that we can obtain up to a 30% improvement in relative response times.