I have experienced the same problem, and run some tests that confirm that the slowdown is related to external monitor resolution. Parallels' responsiveness when using multiple external displays with separate spaces seems to be related to the total resolution of the combined displays.
My configuration is a maxed-out 2016 MBP with Touch ID, running Win10 on latest Parallels (12.1.3 #41532). I devote 12GB RAM and 8 CPUs to my VM. I run with the MBP closed and two external monitors (one running at 3440x1440 and the second running at 2560x1440). My Parallels optimization settings are set to "Faster Virtual Machine", "Better Performance", and "No resource limit". Display are using "Use all displays in fullscreen" and on the MacOS side I'm using "Displays have separate spaces".
When I swapped out the second (2560x1440) monitor for a new 5k Display, I noticed that when I clicked and dragged a window on the secondary display, there was a long pause (~1.5 seconds) before the window would move. After this initial pause, subsequent drag attempts on the same display were immediate (no delay). However, when I then tried to click-and-drag a window on the primary display, the same initial 1.5-second delay occurred. This was repeatable: it seems that whenever the active window changes from one display to another there is a long initial delay. In addition, Parallels had a higher host CPU usage with the larger display connected.
I then switched back down to the original display, and the delay was much shorter (less than one second). I then reconnected the 5k display and the delay again increased. I then connected an even lower-resolution secondary display than the original 2560x1440 and the delay was much less than 1 second.
Unfortunately this delay was unbearable for my day-to-day software development work, so I had to ditch the new 5k display and go back to the old one.
It's worth noting that this delay does *not* occur under the host Mac OS operating system, only when running under the VM in Win10. Also, I've tried the "devices.usb.mouse=0" startup setting mentioned in a KB article and earlier in this thread, but that did not change the behavior at all.
Can anyone from Parallels comment on this? Is it a known issue? Is it being worked on? What should our expectations be in regards to this issue?
Last edited: Jan 30, 2017