The model works pretty well up to much larger systems than you'd expect.

If a particular interrupt has a hard real time constraint, it sounds like a great candidate for a higher priority interrupt which will let it meet that requirement.

The biggest constraint is that this is really a single core model. You need something different if you go to SMP. Though there, AMP where the main core runs this 'interrrupt controller is your scheduler' scheme, and the other cores run against a work stealing scheduler for compute bound work items still is a very nice system to program against.