Looked to me like the skyraider pilot lost sight of the mustang and pulled into him after the mustang broke out of the formation. The mustang broke out and then rolled out of his turn, and the skyraider turned and flew into the mustang. That's what it looked like to me.
Now that it's not "speculation", here's my opinion based on the crash video and accident report.
I find it amazing that the skyraider pilot though it was "normal" to lose sight of the lead aircraft during the pitchout. The second aircraft to pitch out should have become the navigation lead aircraft in order to be able to maintain sight, while the third to pitch would transition his sight from the first aircraft to the second one when the first one pitched out, then follow the second one when it was his turn to go.
That way nobody would lose sight at any point. Terribly poor pre-flight planning, poor flight leadership, poor execution of wingman responsibilities, poor briefing, and the evaluator that checked out the skyraider pilot for the formation demo ought to get a spot evaluation of his own, focusing on formation procedures and lead/wing responsibilities including planning, briefing, in-flight, and post-flight debrief. There is no excuse for this collision. The skyraider's own statements show that he had no business being in that formation, and the mustang driver should have been very concerned about having him on his wing. Losing sight routinely during a pitchout of this sort is stupidly poor planning and execution, and it could have easily been planned, briefed, and flown in such a way as to allow the skyraider pilot to maintain visual through the entire maneuver. Stupid.
My last bit of opinion is that had the plane felt more flyable, a reasonable thing for the mustang driver to do would be to climb up a few thousand feet and do a controllability check. This is standard USAF procedure whenever structural damage is known or suspected. From a high enough altitude to lose control and still safely bail out, the plane is gradually slowed and configured for landing, until the plane either becomes very difficult to control or a reasonable landing speed is reached. If the slowest speed and configuration is unacceptable for landing, then the plane is flown to a bailout area and the pilot bails out. If the plane is controllable at a reasonable landing speed and the landing gear can be lowered properly, then the pilot returns to the nearest suitable runway (preferrably a long/wide one) and lands in front of a large crowd of emergency response personnel.
We had a midair between 2 T-37s about 9 years ago, here at Sheppard AFB. One of the T-37s lost a wing and that crew safely ejected. The other one had its nose section shoved up badly enough to make lowering the landing gear a very bad idea, so after a controllability check done above 10,000 ft and with a chase ship watching things from a safe position, they returned for a belly landing on our biggest runway.
The difference of course is that military fighters and bombers have ejection seats. Controllability checks in non-ejection seat aircraft are a bit more hairy, doubly so if the crew does not have parachutes. But the procedure still applies... Climb to a safe altitude, and then see if you can get the plane configured and slowed down enough to land before it becomes unflyable. In the case of this midair between the mustang and skyraider, obviously the stang driver didn't think the plane would fly long enough to give it a try so he quite reasonably bailed out. He hit the tail on the way out due to the plane pitching down when he released the controls, which is one of the big hazards of hopping out of planes like that. It's easy to second guess his bailout technique since the WWII standard for such a situation seems to be (based on historical accounts I've read) to climb up, slow down, jettison the canopy, roll inverted, release seat belt and fall free. But that assumes that the plane is flyable enough to do all that.