I have a bunch of systems running R80.40 and R81.10, both installed clean (with XFS) and upgraded from lower versions (so still running ext3). Most use spinning disks, so should be pretty much the worst case in terms of storage performance on dedicated hardware. I take snapshots live all the time with no issues.
A snapshot on GAiA seems to involve taking an LVM snapshot (to ensure filesystem consistency), creating a new LV in unallocated PV space, copying the data from the snapshot to the new LV, then deleting the snapshot. It's extremely disk-I/O-intensive, but shouldn't affect the processor or RAM load noticeably.
While the LVM snapshot is live, all reads and writes from the normal filesystem would be multiplied, since XFS and ext3 aren't actually aware of the snapshot. This is one of the reasons I like ZFS so much for my personal systems. Since ZFS is both the volume manager (handling arranging the data on the physical storage) and the filesystem (handling the mapping from file names to data blocks), and it's copy-on-write, volume-level snapshots are extremely fast to create (milliseconds) and don't have an ongoing performance impact while live.
Virtualization would also hurt disk I/O performance pretty seriously. Maybe the combination of the virtualization hit and the snapshot hit causes some disk operation during a policy push to time out? Are you using spinning disks or SSDs for the VM datastore?