Bear in mind that's a conceptual description. The read/write barriers are implemented as compiler node graphs and thus can be optimised.