If we live in a world where reality is on deck, and not the seemingly arbitrary and capricious one that folks sometimes weave, I think you can determine what answer is "better" with more context.
Renting a bus and filling it with golf balls is significantly more effort intensive than calculating, but it might give interesting, real world insights if the values differ.
For the purpose of running a business, the choice, to my mind, hinges on how much your revenue / customer satisfaction / product depends on the real world intelligence such an endeavour might bring. The situation illustrated is contrived, so it's hard to contrive an answer to that question.
There's hybridized approaches. Here's a 3D/AV interactive developer's contribution to your approach list:
I've used lidars and depth cameras as a hammer quite a bit over the last year, so I'd do thus - Realsense D435 capture the inside of a bus. Use meshlab to align and stitch all your pointsets. Perform reconstruction on it so you get a 3d mesh. Use your tool of choice to perform a physics simulation on golf balls (I'd probably use Unity3d, because customers love simulations they can play with.)
Run your simulation a ton of times with epsilon changes to initial conditions. Maybe even do it with an evolutionary algorithm that tries to maximize the count!
Either way, report back with a distribution. Also capture frames every once in a while so you can give them a timelapse. God customers love timelapses.
