I believe that this was done to keep performance optimal on inserts/updates. If you could create a UPI that did not include the PPI column, then each insert (or update of the primary index columns) would have to go through each partition to make sure that the primary index values for the row were unique. This would be a lot of overhead for each insert and/or update of the PI columns.
You can create a USI on the unique columns if you want to enforce this. Since the USI subtable is arranged in hashing order on each AMP, a duplicate row would be detected most likley in the same data block.