A large number of A/C units potentially represents a substantial flexible load susceptible to be leveraged for Demand-Side Management (DSM) purposes. However, in such a scenario, a drop in quality of service (QoS), i.e. an uncomfortable increase in temperature, is to be expected as some cooling-down is delayed to achieve peak reduction. We present the results of a simulation study designed to test a decentralised control framework based on probabilistic activation of individual A/C units and reactive pricing [1]. Performance is evaluated and discussed, primarily in terms of the trade-off between QoS and DSM efficiency.