As measured by MSE, bias and noise are independent and additive sources of error. Obviously, bias is always bad and reducing it always improves accuracy. Less intuitive is the fact that noise is equally bad and that reducing noise is always an improvement.
