This paper focuses on robust compensation of delays within a multi-input multi-output discrete-time feedback loop for application to real-time distributed control systems. The delay compensation algorithm, formulated in this paper, is an extension of the standard loop transfer recovery (LTR) procedure from one-step prediction to the general case of p-step prediction (p≥1). It is shown that the steady-state minimum-variance filter gain is the H2-minimization solution of the relative error between the target sensitivity matrix and the actual sensitivity matrix for p-step prediction (p≥1). This concept forms the basis for synthesis of robust p-step delay compensators (p>1). The proposed control synthesis procedure for delay compensation is demonstrated via simulation of the flight control system of an advanced aircraft.