We completely agree about predictability; clearly you understand how important this is, and how/why a strict physics-based approach produces it, but that is not something that is obvious to many, until they have actually experienced it. I attended siggraph and various other conventions for NL, and this was always one of the first things I’d try to explain to people.
As far as quality goes, even maxwell will not be able to compete with bella. Everything in bella is brand new, designed with the benefit of so many years of hindsight, and without the baggage that you end up with in a codebase that’s been maintained for nearly two decades. Our scene model, material model, sky model, camera model — everything — is simply superior. Not in a bragging way, but just factually. An architect can easily understand this — just imagine you had worked on a single skyscraper for 15 years, and how you would wish you could go back and start over; there would be entire methods of construction that had not even existed when you started, but it would be too late to make use of them.
Our disadvantage is simply that bella is still young, and therefore still lacks various features (e.g. hair/grass, volumetrics, etc). But those are just a matter of time, not capability. Luckily, being all brand new & modern, we are able to implement such things much quicker and better than would otherwise be possible. We invested a good deal of time & effort to design our foundation, and it has been paying off already, and will continue to do so.
Unfortunately though, our gallery does not yet reflect what bella can really do, since it is still the case that most of the images have been made by me, and besides not being a great visualizer, it is difficult to stop coding long enough to spend the time necessary (as you well know — i’ve seen your outstanding work) to produce really great imagery.