HVAC systems use a substantial part of the whole energy usage of buildings. The optimizing of their operation can greatly affect the power use of a building, making them an interesting subject when trying to save energy. However, this should not affect the comfort of the people inside. Many approaches aim to optimize the operation of the heating and cooling system; in this paper, we present an approach to steer the heat pumps to reduce energy usage while aiming to maintain a certain level of comfort. For this purpose, we employ a market-based distributed method for power-balancing. To maintain the comfort level, the market-based distributed system assigns each device a cost-curve, parametrized with the current temperature of the room. This allows the cost to reflect the urgency of the HVAC operation. This approach was tested in a real-world environment: we use 10 heat pumps responsible for temperature control in 10 comparable-sized rooms. The test was performed for 3 months in summer. We limited the total peak power, and the algorithm balanced the consumption of the heat pumps with the available supply. The experiments showed that the system successfully managed to operate within the limit (lowering peak usage), and - to a certain point - reduce the cost without significantly deteriorating the working conditions of the occupants of the rooms. This test allowed us to estimate the minimal peak power requirement for the tested set-up that will still keep the room temperatures in or close to comfortable levels. The experiments show that a fully distributed market-based approach with parametrized cost functions can be used to limit peak usage while maintaining temperatures.