If an XSUB were used to make the conversion ukeysort would be the fastest!

And to show it, I have just uploaded Sort::Key::IPv4 to CPAN ;-)

Now, adding this test to the benchmarks...

use Sort::Key::IPv4 qw(ipv4sort);

sub ipv4ks {
    my @sorted = ipv4sort @address;
}
[download]

I get for 10_000 addresses:

               Rate gloryhackish     ukeysort      keysort     ipv4sor
+t
gloryhackish 6.14/s           --         -31%         -54%         -79
+%
ukeysort     8.91/s          45%           --         -34%         -70
+%
keysort      13.5/s         119%          51%           --         -54
+%
ipv4sort     29.4/s         378%         229%         118%           -
+-
[download]

For 100_000:

             s/iter gloryhackish     ukeysort      keysort     ipv4sor
+t
gloryhackish   1.80           --         -31%         -33%         -78
+%
ukeysort       1.25          44%           --          -3%         -68
+%
keysort        1.21          49%           3%           --         -67
+%
ipv4sort      0.400         350%         212%         202%           -
+-
[download]

And for 1_000_000:

             s/iter gloryhackish      keysort     ukeysort     ipv4sor
+t
gloryhackish   24.3           --         -29%         -46%         -79
+%
keysort        17.2          41%           --         -24%         -71
+%
ukeysort       13.2          85%          31%           --         -62
+%
ipv4sort       4.99         388%         245%         164%           -
+-
[download]