This paper quantifies the value of a large population of heterogeneous thermostatically controlled loads (TCLs). The TCL dynamics are regulated by means of an advanced demand side response model (DSRM). It optimally determines the flexible energy/power consumption and simultaneously allocates multiple ancillary services. This model explicitly incorporates the control of dynamics of the TCL recovery pattern after the provision of the selected services. The proposed framework is integrated in a mixed integer linear programming formulation for a multi-stage stochastic unit commitment. The scheduling routine considers inertia-dependent frequency response requirements to deal with the drastic reduction of system inertia under future low-carbon scenarios. Case studies focus on the system operation cost and CO2 emissions reductions for individual TCLs for a) different future network scenarios, b) different frequency requirements, c) changes of TCL parameters (e.g. coefficient of performance, thermal insulation etc.).