A method for communication in a network that includes multiple nodes having respective network interfaces and interconnects between the network interfaces, which include at least first and second network interfaces connected by a physical interconnect having a given latency. The method includes defining a target latency, greater than the given latency, for communication between the first and second network interfaces. Data are transmitted between the first and second network interfaces over the physical interconnect while applying, by at least one of the first and second network interfaces, a delay in transmission of the data corresponding to a difference between the target latency and the given latency.