With the increasing complexity of distributed control tasks based on their intellectualization, there are problems of insufficient time and computing resources for functioning in real time. In this regard, there is a need to develop methods for organizing distributed real-time computer systems, based on the consolidation of distributed computing resources with their integration into clusters. The possibilities of increasing the probability of timely servicing of waiting- critical requests in the cluster as a result of query replication and controlling the time of destruction of potentially expired replicas in node queues are investigated. The cluster is represented as a group of queuing systems with infinite queues with a limited average waiting time. The effectiveness of the reserved service of a real-time request is determined by the probability of executing at least one of the generated copies of the request in the maximum allowable time without losing it due to errors and waiting time limits in the queues of cluster nodes. It is shown that there is an optimal multiplicity of query replication with a significant influence of the choice of restrictions on the waiting time for requests in queues before they are destroyed.