He made a minor mistake. In his high estimate, he used the high end of the range for de-duplication savings (20%) when he should have used the low end (5%). Not a big error (something like $4.5m/mo instead of $4m).