This study investigates how constraint-based routing decision granularity significantly affects the scalability and blocking performance of QoS routing in an MPLS network. Coarse granularity, such as per-destination, has lower storage and computational overheads but is only suitable for best effort traffic. On the other hand, fine granularity, such as per-flow, provides lower blocking probability for bandwidth requests, but requires a huge number of states and high computational cost. To achieve cost-effective scalability, this study proposes using hybrid granularity schemes. The overflowed cache of the per-pair/flow scheme adds a per-pair cache and a per-flow cache as the routing cache, and performs well in blocking probability. The per-pair/class scheme groups the flows into several paths using routing marks, thus allowing packets to be label-forwarded with a bounded cache.