It would be great if we could have the mean difference as being (2) - (1) / (3) - (1) and (3) - (2). As (1) is usually the control group, I think it makes conceptually more sense to have treatment - control than the opposite to have a positive result when treatment is effective.