A partially decentralized model reference adaptive controller proposed earlier is modified for improving the bad zero-state performance that may result from the poor knowledge of the subsystem parameters. The improvement in performance is achieved by treating the parameter mismatch in each local subsystem as an external disturbance and then designing an auxiliary local feedback to reject the effect of this mismatch on the local tracking error. The resulting closed loop system is analysed in the presence of unmodelled interconnections and it is shown that, whenever the uncertainty in the high frequency gain of the transfer function of each isolated subsystem is less than 100%, the modified scheme results in improved zero-state performance provided that the unnodelled interconnections are ’small enough’. Furthermore, the performance improvement, achievable using the proposed scheme depends on the size of the unnodelled interconnections in such a way that, as the unmodelled interconnections disappear, the improvement in zero-state performance can be made arbitrary. The proposed scheme uses high gain compensators in each local control law and so sensor noise amplification is a potential trouble-spot.