This paper focuses on the control problem of networked systems with Markovian delays existing on not only the feedback channel but also the forward channel. The resulting controller is thus two-mode dependent by making full use of the available delay information on both channels. Unlike the existing results by using the augmentation method for incorporating the two-channel Markovian delays into the closed-loop model, a prediction scheme is proposed to design a sequence of all possible two-mode dependent controller gains by efficiently solving the derived linear matrix inequalities and send them to the actuator. Depending on the actual delay on the forward channel, the appropriate controller gain will be implemented to drive the plant. Furthermore, the unavoidable uncertainties of the elements in the probability transition matrices are considered in the controller design. Simulation studies and comparisons demonstrate the effectiveness of efficiency of the proposed method.