This work presents a new control approach to multi-contact balancing for torque-controlled humanoid robots. The controller includes a non-strict task hierarchy, which allows the robot to use a subset of its end effectors for balancing while the remaining ones can be used for interacting with the environment. The controller creates a passive and compliant behavior for regulating the center of mass (CoM) location, hip orientation and the poses of each end effector assigned to the interaction task. This is achieved by applying a suitable wrench (force and torque) at each one of the end effectors used for interaction. The contact wrenches at the balancing end effectors are chosen such that the sum of the balancing and interaction wrenches produce the desired wrench at the CoM. The algorithm requires the solution of an optimization problem, which distributes the CoM wrench to the end effectors taking into account constraints for unilaterality, friction and position of the center of pressure. Furthermore, the feedback controller is combined with a feedforward control in order to improve performance while tracking a predefined trajectory, leading to a control structure similar to a PD+ control. The controller is evaluated in several experiments with the humanoid robot TORO.