Follow the best practice of having a zero baseline? Or emphasize small (but practically significant) differences by changing the baseline?

Why not both?