The examples you mention work well because there is a real world affordance (the wheel, the act of throwing) that the gesture maps to. I agree that rotating a wheel gesturally makes more sense as an interaction in VR than the same action does with a mouse pointer.

The grenade example is a good one, because throwing/flicking doesn't require much feedback, unlike say rotation or fine grained manipulation. Of course the interaction would feel nicer if you could feel the weight of what you're throwing - the physics of throwing objects always feels kind of off in VR games - but it still works well enough. Note that many games did this with the Wiimote, and there is nothing about that interaction that is singular to VR.

However, the moment you lose that affordance based in a real world object, and ask your user/player to make gestures in mid air based on recollection, things really do start to fall apart. There is a host of HCI literature/projects around this - e.g. check out papers/demos from the Kinect era. An RPG game like the one you describe could be made with Kinect, and I vaguely recall a few in development - but Kinect demonstrated well that the only kind of gestural interaction that works in long term sessions is dancing.

Btw, this is why if you go to e.g. the VR center in Tokyo, almost every booth has physical props for you to hold/grip/sit on while you're fishing/golfing/shooting a gun/riding a bike in VR.