Consider the following scenario:
We have multiple customers with mixed environments of Server 2008 R2 and Server 2012 R2 VMs. To access those servers for administrative purposes, we VPN to each customer's site with Sonicwall Global VPN clients, then RDP to each machine.
When we RDP to VMs running Server 2008/2008 R2, session performance is normal, and disconnections/hangs are rare. When we RDP to 2012/2012R2 VMs with equivalent hardware resources, on the same network, and sometimes even on the same hypervisor, the RDP connection quality meter shows only one or two "bars", sessions are sluggish, and disconnections are constant. Sometimes I have to reconnect to the same VM 6 or 7 times in a ten minute period.
Here's where it gets interesting: if you're actually on the customer's network (say by sitting at one of their PCs) and RDP to the 2012 VMs having trouble, session performance is perfectly normal. In other words, it's only when VPN access is used, that performance suffers. Which suggests something about the VPN or the firewall it traverses is misconfigured, but if that's true, why is RDP session performance for 2008-based VMs unaffected when we VPN in that same way?
As I said, we ruled out hardware resource limitations. I also tried turning off all client redirection features (printers, audio, etc.), and downgrading the color and other 'visual experience' settings, to no avail. And the performance problems occur when RDPing over VPN from multiple Windows PCs, running everything from 7 to 8.1, so it's not just my machine that's having trouble establishing stable sessions.