It looks like my disk access speed testing is to blame for the jumps in the disk usage. The tests created a 1GB file every 5 minutes, and sometimes it came into sync with the hourly ZFS snapshots, resulting in jumps in the size of the ZFS snapshots over the past two weeks.

We have manually destroyed some early May snaps on h1fs0 (and h1fs1) which has freed 26GB immediately and will free up more space as the large snapshots roll off the look backup period.