Anyway, after reblessing the object, wouldn't it be better to take the reference in numeric context than to extract the number from the string form?

True, although I've no idea off the top of my head whether that behaviour has always been in Perl so there may be issues with the perl versions that Scalar::Util supports.

You might be right... the phrase

Using a reference as a number produces an integer representing its storage location in memory. The only useful thing to be done with this is to compare two references numerically to see whether they refer to the same location.

has got into perlref between 5.005 and 5.6.0. The feature might have been there from the start though.

If it's backwards compatible, and if speed is the utility function, yes. Surprisingly, unless I'm missing something, it's even faster than refaddr, at least for my environment (perl 5.8.6 on linux i386).

Update: clearly, what I'm missing is reading the docs to Benchmark and proofreading my code. Stupid errors fixed, I hope. 0 + $ref still seems to be faster than the XS refaddr, which is surprising, but maybe that's the function call overhead.

use strict;
use warnings;
use Scalar::Util qw( refaddr );
use Benchmark qw( cmpthese );

my $ref = {};

cmpthese( -5, {
    'xs   ' => sub { refaddr $ref },
    'regex' => sub { "$ref" =~ /0x(\w+)/; hex $1; },
    '0+ref' => sub { 0 + $ref },
});
[download]

Result:

           Rate regex xs    0+ref
regex  165816/s    --  -89%  -97%
xs    1504549/s  807%    --  -73%
0+ref 5550057/s 3247%  269%    --
[download]