The concept of HPC as a Service (HPCaaS) aims to shape traditional High Performance Computing (HPC) practice into the model of cloud and services computing. Two important concerns of HPCaaS are easier access to HPC facilities and efficient parallel job scheduling. Moldable job scheduling has become an important research issue since it can effectively deal with the two important concerns of HPCaaS. This paper presents a new moldable job scheduling approach based on a two-level utilization mechanism. A series of experiments have been conducted to evaluate the proposed approach and the experimental results indicate that our approach can achieve up to 22% performance improvement in terms of average turnaround time, compared to the previous method in the literature.