This paper deals with the problem of the loss of performance of the pair observer-controller, when measurements have a delay due to communication over networks. Here we consider the case where the estimation of the states is carried out using a moving horizon estimator (MHE), the control actions are computed by using a centralized model predictive controller (MPC), and the delay varies randomly and is n times the sampling time (n ∈ ℕ). In order to tackle the loss of performance associated with the pair MHE-MPC, an MHE with variable structure is proposed. The resulting pair MHE-MPC was tested using the four tank process as a test bed showing an improvement on the performance.