It really depends. If you e.g. allocate items inside a loop to build up an array to later do something on that, you'd want defer to run only for the allocated items (and thus the statement needs to be inside the loop body), but deallocation should only happen once the array is actually used.

If we're going to have that kind of semi-manual resource management, anyway, it would be better to have labelled blocks and the corresponding defer-like statement that can reference them. Something like:

   outer: {
      ...
      inner: {
         defer close(x) after outer;
      }
   }