I've seen a lot of these with VSC 4.2.2/cDOT 8.2.2/ESXi5.5.
Unfortunately they're intermittent / irregular so we've not been able to determine a pattern enough to file a workable bug.
There's definitely some corrolation to it hitting busier VMs, and shifting other workloads to compensate for the vmdk-datastore snap has alleviated some of it.
If you can manage to get a BURT opened, post it, I'll definitely tag against it.