Option C is the best choice because it directly leverages Delta Lake's partitioning feature to efficiently isolate and manage data based on the partitioning schema (year and month). Dropping entire partitions that contain data older than 3 years is a highly efficient operation that minimizes the impact on query performance for current data and complies with data retention policies. Option A is inefficient as it involves scanning and deleting individual records, which is time-consuming and resource-intensive. Option B is incorrect because the VACUUM feature is not designed for selective data retention based on age; it removes files not referenced by the Delta table, which could include recent data if not used carefully. Option D, while feasible, is more manual and less efficient than dropping partitions, especially for large datasets.