Fail in what way? In not producing the exact (save for whitespace) copy of the original code? Is that really a problem? Both the compiled and "deparsed" code would still have to denote that the subroutine isn't returning the value of the statement before the empty return.

Empty return is the only solution that is NOT a syntax trick, therefore it ought to be implemented efficiently and if B::Deparse needs to produce code of a subroutine that returns nothing, the empty return should be the "inferred" solution even if it means that deparsing turns some of those tricks into an explicit empty return.