Thinking about it some more made me realize that ref $x eq 'REF' can be used more efficiently then ref $x eq 'SCALAR'. (Although not really on the previous code, since the goal was to keep one reference at the end)

If you have ref $x eq 'REF' then you know that the scalar to which $x refers to is yet another reference, so noone stops you from doing $x = $$$x;, which avoids yet another call to the ref function...

But then you are doing two calls to ref instead of one...

But then another question would be, which one would be the most efficient? (assuming the perl-interpreter does not return REF (and doesn't do any check on it))

The current behavior would be the more efficient one. But it is also less clean, conceptually, so (I'm come to conclude) it shouldn't belong in the ref builtin.

This doesn't matter, though. It is a *very simple extension* to add this feature outside core perl. Something along the lines of:

use Inline C => << "EO_REFREF";

int
refref(SV* thing) {
    return SvROK(SvRV(thing));
}
EO_REFREF
[download]

(This is not anywhere near tested, though :)