However even so I would argue against turning it on naively.

Oh I agree. It's not something that can be globally switched on for perl 5 (and the lousy speed for subroutine calls makes tail-recursive code less appealing anyway). I just don't think caller is a problem :-)

When the things destroyed are objects that may have side-effects in their DESTROY methods (see ReleaseAction for an example where this happens) then trying to optimize tail-recursion creates a significant change in what the code does.

True, but naive scope-related DESTROY code is going to be in trouble when we get proper GC in perl6 anyway :-)