React and Imba can both optimize using shouldComponenUpdate (in Imba you would add the same logic inside node#commit). Utilising shouldComponentUpdate (in React and Imba) requires additional logic by the developer, not the framework. This is not an optimisation at the framework level - but an endpoint where the developer can write their own custom optimisations. Adding your own manual checks all around your app to check for statechanges is inelegant, cumbersome, and bug prone. In my opinion it defeats much of the purpose of a virtual dom, if you are expected to do your own diffing before sending it through. Also, the relative performance difference is about the same when turning it on.
I fully stand behind the benchmark, and would even go as far as saying that it is more relevant for real world applications than benchmarks like dbmonster where everything in the view updates on every render. This effectively tries to calculate how quickly each contestant can reconcile the whole view (synchronously), but when only parts of the view has actually changed.
