> but in that case they could just release a second dataset as an addendum

Or for some data it would make sense to partition the data into smaller chunks instead of one huge archive. That way adding a chunk (the new year's data for a multi-year dataset perhaps) just menas releasing a new torrent with the extra srchive in and a name meaningful enough to indicate the difference. Anyone with the last set could then just download the new partition (and any modified ones).