improve diff-delta with sparse and/or repetitive data