@teohhanhui@slp I think it should be set to 80% or so, yeah, though this is iffy until some change is made to bypass the guest page cache ^^ (Sergio said it's coming)
There's also a GPU memory leak in krun, but that accidentally helped me test memory pressure and the VM seems to fairly gracefully get swapped out/zram compressed when it's too much so I think it will work ^^
I also mentioned to Sergio that we should definitely look into a client-server architecture so there is only ever one krun VM even if you launch multiple apps from the host, since multiple VMs competing for RAM is probably a bad idea...