The hypervisor would have to envelop the entire OS to be tight and effective, wouldn't it? Like the famous
Blue Pill...
Then, using OS structures, a single process could be singled out for debugging, so the experience is not too slow.
Indeed, with a network connection a timing based anti-debugging technique could be made.
What about, non-timing based techniques?
Could IO MMU and hardware be exploited?