Your wording suggests you'd be using ACIM or PMAC motors, with a separate power stage for each side; For DC motors, you can do it with one controller.The differential is simulated by the controller, which takes into account
the turning wheel state and accelerates the wheels differently.
Turning without throttle should not pose a problem since the wheels
are all independent (hence, no differential required --> I suppose ?)
I want to know if this would be reasonably safe (for a RWD) and worth
mentioning, or does it have some serious flaw ?
The formulas linking steering angle, wheelbase, track, turning circle and ideal speed for each wheel are all pretty basic geometry.
The problem with implementing a "software diff" is that your throttle is normally configured as a torque request rather than a speed control, but you'll want to reduce torque to the inside wheel if the weight transfer causes it to lose traction. You'll need a clever algorithm to tell the difference between acceleration and wheelspin.
A basic system would be to use steering angle to determine which is the outer wheel, apply torque control to that wheel (as in a single-motor system), then derive the inner wheel's target speed from the measured outer wheel speed.