It sounds like this is column level partitioning.Each column or columns (based on partitioning expression) is stored as different subtable (or something similar) on disk.If only few columns are frequently accessed, they can be put on cache/faster disk or other neat optimizations for join processing.