>This seems mostly useful for laptops with both integrated and discrete GPUs but no extra monitor output and for non-fullscreen 3D on desktops, since for fullscreen 3D on desktop you can just run a second cable to the monitor, which will be faster and allows the guest to drive monitor sync.

i've been doing it for years (albeit with qemu), even with additional monitors available.

It's nice to keep your workflow where you want it. I use a unix-alike OS nearly always, but work will inevitably pull me into a windows software that'll bitch about GPU acceleration (say.. autodesk fusion) even if it doesn't need it for the task.

It's nice to keep that window where I want it. Sure, I loses sync/cec stuff, but I can drag that window across my entire workspace of monitors, and that's pretty powerful when you're stuck in a VM for a single piece of software.

When I use a VM for gaming, that's when I run the GPU by itself to a separate monitor. Gaming VMs are ran weirdly anyway -- it'd probably have access to an entire block device rather than some VDI equivalent, PCI access, etc. That's more like just running two computers off of one mainboard at that point; not much to keep isolation sanitary.