Networked control systems (NCS) consist of sensors and actuators that are connected to a controller through a packet-switched network in a feedback loop to control physical systems in diverse application areas such as industry, automotive, or power infrastructure. The control of critical real-time systems places strong requirements on the latency and reliability of both the communication network and the controller. In this paper, we consider the problem of increasing the reliability of an NCS subject to crash failures and message loss by replicating the controller component. Previous replication schemes for real-time systems have focused on ensuring that no conflicting values are sent to the actuators by different replicas. Since this property, which we call output consistency, only refers to the values within one time step, it is insufficient for reasoning about the formal conditions under which a group of replicated controllers behaves equivalent to a non-replicated controller. Therefore, we propose the stronger state consistency property, which ensures that the sequence of values produced by the replicated controller exhibits the same dynamical behaviour as a non-replicated controller. Moreover, we present SCRaM, a protocol for replicating generic periodically sampled controllers that satisfies both of these consistency requirements. To demonstrate the effectiveness of our approach, we evaluated it experimentally for the control of a cart-driven inverted pendulum.