For a columnar database, this would probably work fairly well as it would put data of similar compressability adjacent to more of its kind, but more row-based databases where various types of data is intermingled would probably not benefit nearly as much.