What this means is that if you have a scene that is tone-mapped to just fit within the dynamic range of the monitor, and then the brightest lights in the scene are cut, you have a choice: either use a tone mapping that is similar to the previous frame, resulting in an apparently under-exposed frame that uses only a fraction of the dynamic range of the monitor, or use a radically different tone map, which would replace the sudden darkening effect with colors shifting all over the place.
A naive automated tone-mapper might take the latter approach in order to be embarrassingly parallel, but in order to look natural, the tone-map would have to shift gradually, just like pupils dilate gradually. These inter-frame data dependencies will make tone-mapping video comparable in complexity to video encoding. It also means that it is unlikely to be a task that can be fully automated, because there is a tradeoff between inter-frame contrast and intra-frame contrast that can only be resolved with knowledge of the artistic intent of the sequence.