Neatly removes the lhs/rhs distinction in the compiler, although in practice tracking during the translation whether evaluating e.g. "a[i]" needs to end with DEREF or not ain't that onerous.
But "The Design of an Optimizing Compiler" by Wulf et al. is an interesting read nonetheless; especially if you wonder how people did register allocation before Chaitin's seminal paper.