Because of the somewhat pathological benchmark design, he could probably implement a component for each picture and set shouldComponentUpdate to simply return false. With proper keys, I have a feeling the React performance would improve dramatically, perhaps as much or more as the pureRender mixin.