I was doing index rebuilds online (yesterday) and had to cancel them due to blocking. I think that's the reason for the extra sys.partitions records. There wasn't an associated record in sys.dm_db_partition_stats so I did a right join to show that they're orphans. I was able to do the rebuild offline, but that didn't clear it up. I was right about to do another index rebuild online, but I wanted to query the sys.partitions again and what do you know... It cleaned itself up. I didn't do anything at all.

I am really curious why it took so long to clean itself up though. I was trying to replicate the issue on a backup and I was able to get the extra sys.partition record. However, it cleared itself up in about a second. I wonder if it's because that table is constantly being queried and there just happened to be a lull so it could clean itself up properly?