In this paper, we study the use of redundancy to reduce latency and increase reliability in communication networks. The work is motivated in particular by wireless networks, where channels may be unreliable and channel characteristics may vary over time. The objective is to provide an analysis of the gains achievable through the use of redundant messages. Redundancy increases the network load and hence the delay of each packet, but can reduce overall delay by exploiting independent randomness across multiple paths. One of the goals addressed in this paper is the optimisation of this trade-off. We consider both average delay minimisation and probabilistic guarantees on delay exceeding some tolerance threshold. We study a number of different stochastic models for packet transmission times, and uses techniques from queueing theory and large deviations to analyse their performance. The analytical results are complemented by simulations.