This is precisely what happened when the US Navy F-16Ns were used exclusively by TOPGUN, and had a short service life before the airframe developed cracks and went to the boneyard. These jets were flown harder than the USAF flew them in operational service.
Also, the aerodynamic control surface effectiveness on fighter jets does not allow rapid reversals at some magic level significantly beyond what a human pilot can handle. An old A-4 Skyhawk can roll beyond 360 degrees a second with a human pilot.
Another issue is that thrust vectored aircraft like the F-22 Raptor, will rapidly loose airspeed at a relatively low sustained turn rate. This turn rate is way less than the 9G maximum of the aircraft. Additionally, An F-22 (91-4003) was effectively written off after a -11G overstress mishap. The pilot was okay, but the jet never flew again. To fly a fighter in an expanded G envelope would require additional weight, negating some of the benefits of unmanned fighters.
the designers are targeting a certain g-limit assumed for the pilots, and then make the plane as light as possible.
perhaps we will see hybrids in the intermediate term : humans providing guidance to the unmanned craft (with latency and jammability) and the craft autonomously doing the low latency fighting. i'm not sure this is "worse" if all the planes are unmanned. granted any fighting is bad.
this also has big implications for design -- reliability requirements, disposability change. one way missions.
the comments re: once the virtual fight ends the humans start fighting, i don't think so. they won't be as good, else the ai version wouldn't be in service, so that would be pointless.
If the plan was for smaller, miniature, one-way drones, then something like a ADM-160 MALD or AGM-154 JSOW would suffice, but it wouldn't be a fighter sized drone.