The mental model is much simpler because your `render` method describes UI at any point of time. Not just the DOM, but your own components too. On the inside, any component can have its own state, but from the outside, it's not visible, and components using it don't care or know about it.
Separating `props` and `state`, component boundaries, lack of two-way binding and predictable top-down data flow make it easy to reason about where any data comes from, and how UI will change over time.
This is true of all virtual dom libraries. There is no point of two-way binding when you have a virtual dom, and I'm not sure it even makes sense since you have to rerender the tree anyways.
So we agree, you're just separating the concept from the implementation and I'm talking about them as one.
There's a difference between “virtual DOM” and components.
In React, components are not just functions that return their own virtual DOM. AFAIK for many vdom-based libraries this statement wouldn't be true.
React components may have local state (as much as some people hate it, some find it useful), they have a lifecycle, can react to receiving new props with side effects, can implement diff bail-out hook. And you can nest such components declaratively.
The other libraries intentionally left out shouldComponentUpdate because they don't think it's something a developer should have to worry about. I don't know of any evidence to suggest that React is faster because of this feature.
And declarative nesting is a feature of all of the frameworks I've come across, I'm not sure why you think that's unique to React. The advantage feature I would credit React for is the size of its community and influential advocates like you.
WPF (even completely ignoring XAML), JavaFX both support declarative nesting with interesting databinding features to support it. They are based on retained scene graphs, however.
If you are interested in academics, I published such a system at ECOOP in 2006:
