According to A modern eye on ML type inference
Furthermore, for some unknown reason, W appears to have become more popular than J, even though the latter is viewed—with reason!—by Milner as a simplification of the former. Jones’ presentation of type inference for Haskell  does rest upon an implementation of J.
Indeed, I've seen many expositions of W. But this is the first one of J that I've seen.
So does W have advantages which these notes overlook? Or is it just about avoiding mutability in Milner's presentation of J?