In this note, we develop performance bounds on disturbance power reduction for multiple-input multiple-output networked feedback systems via an information-theoretic approach. The bounds are derived for control systems with linear time-invariant plants and causal stabilizing controllers communicating over noisy communication channels. We utilize the notion of power gain to measure the worst case power reduction from the disturbance to the error signal. Concepts from information theory, such as entropy and mutual information, are instrumental in the analysis, and the performance bounds can be quantified explicitly in terms of the channel blurredness and the plant unstable dynamics.