Would it be possible to implement the STM/LDM with clever register renames and defer the actual spill into memory? EDIT: given that superscalar implementations

Would it be possible to implement the STM/LDM with clever register renames and defer the actual spill into memory? EDIT: given that superscalar implementations already implement register renaming logic