High CPU and Webadmin stops responding on Hyper-V requiring reboot

I'm having a pretty identical time to what happened in this thread, and have tried all the advice without any results. 

While that seemed to be underpowered hardware, I don't think that is my issue.

I'm running fully patched Hyper-V 2012R2, on a Dell T410 with 40GB RAM, dual Xeon E5504s, a PERC5 with 6 1TB 7200 SATA drives in RAID 6, with the following VMs using the onboard dual port Broadcom NIC:

2 Windows VMs with 1 CPU/2GB. These are domain controllers doing DHCP/DNS for 6 computers, and authentication for 2 desktops and a file server. While operating they never peak above 20% CPU in the guest, and are generally always below 1GB of RAM.
3 Linux console only Linux VMs with 2CPU, 2 have 2GB the other 4GB RAM. The first two do a whole lot of absolutely nothing 99% of the time. The third downloads for about an hour a day over  HTTPS, but is otherwise totally idle. 
1 Linux console only VM with 6CPU, 16GB RAM. This runs Plex media server which has a very bursty usage and generally maxes itself out (CPU and RAM wise) for about 5 minutes 3-5 times a day. 

Load levels are typically very very close to idle.

My UTM VM has 2CPU and 4GB RAM and a dedicated dual-port NetXtreme II BCM5709. I'm running firmware 9.355-1 and using pattern version 96565.

CPU spikes to 99% of what's allocated according to Hyper-V console, the webadmin takes ages to load and eventually either times out or gives a 503 error. Internet throughput stays at around 250Mbps out of 350Mbps normally. Restarting the HTTPD doesn't fix anything, rebooting the device from shell or by kicking it from the Hyper-V console brings the box back up and everything functions properly for a few days. The drops in the RAM on this graph are where I've had to kick the VM. The times where it starts to spike don't seem to correlate with anything going on on any other VMs or the host.

I've got a home license with ~20 firewall rules and country-blocking for essentially everywhere, ATP and IPS are enabled but I've never configured anything on them, dashboard says Intrusion Prevention is active with 1278/26461 patterns (if that is relevant). Anti-portscan is also enabled, I have 2 IPSec tunnels configured. 

Is it possible that something is getting messed up when the Plex server maxes itself out, and isn't nicely coming back? Other than that, the  UTM VM shouldn't ever feel any form of resource pinch, and even then with the rest of the boxes idle, it should be fine.