This work addresses the consideration of the energy landscape roughness in protein sequence design. The proteins are modeled by 2D lattice chains, initially designed to maximize the energy gap between the folded and unfolded states. Additional optimization and control of the folding properties is achieved by specific sequence mutations that alter the energetic and geometric roughness of the landscape. It is found that mutations that reduce the energetic roughness at the expense of increasing the native-state energy generally lead to a fast folding and stable protein at lower temperatures. Such mutations are also found to modify the geometric roughness (related to nucleation effects) creating variations in the folding time-that depends specifically on each sequence and can lead in many cases to a reduction of the total landscape roughness. An additional reduction of the geometric roughness is achieved by adding local bond-angle propensities to selected sequence sites.