I'll defer to aerohead on the details, but you might start with tuft testing. If the air is rising along the sides, it's moving faster over the top, so then doing just the windows might be enough.
Here's an example (lower left) of using hot air in a Coanda nozzle. VWs produced about 1500cfm of cooling air. I'm not sure a heater core would affect the amount of air that would be moving through the [sub]system.
Can you see some resemblance to the Tubik?