A 2.5 TB sales fact table is queried 90% of the time with predicates on ORDER_DATE and REGION_CODE (approximately 150 distinct regions). The remaining 10% of queries are ad-hoc and use various columns. In which scenario is defining a clustering key most beneficial?