I think he's trying to use numbers to illustrate his point, but he's not declaring anything empirical. When asking “Maybe 20%?” the implication is it is at most 20% better, but certainly not 50% and never 100% better.