We were seeing this in our environment too - and it was actually taking down IIS servers that were serving content from the NetApp CIFS share. I raised a support ticket and they pointed me to the following article, which basically says in 8.3.1 they introduced the ability to manage previously hard-coded thresholds for CIFS/SMB related session parameters:
The default of 100 was not high enough in our environment for max-watches-set-per-tree. We set this to 500 (max value is 4294967295 as per the man page). I've confirmed with NetApp and during implementation the change is non-disruptive to existing CIFS sessions. You do need to execute the command from advanced priv mode.
We're still waiting to validate it's fixed our issue, though I'm highly confident as the issue started immediately after our upgrade to 8.3.1, after 2 years of otherwise stability.