In a simplistic benchmark of a thousand records the straight substr $a cmp substr $b comes out over 180% quicker than the ST version. Using a GRT saves another 14%.

The cost of the split in the ST outweights the repeated substr in this case.

#! perl -slw
use strict;
use Benchmark qw[ cmpthese ];

open IN, '<', 'test.dat' or die $!;
our @lines =  <IN>;
close IN;

print "Sorting ", scalar @lines, $/;

cmpthese( -3, {
    ST => q[ 
        my @sorted = map $_->[0], sort{ 
            $a->[4] <=> $b->[4] 
        } map [ $_, split /:/ ], @lines; 
      ],
    XX => q[ 
         my @sorted = sort{ 
            substr( $a, 19 ) cmp substr( $b, 19 ) 
         } @lines; 
      ],
    GRT=> q[ 
         my @sorted = map{ 
             substr $_, 10 
         } sort map{ 
             substr( $_, 20 ) .  $_ 
         } @lines; 
      ],
});

__END__
P:\test>318176
Sorting 1000

      Rate   ST   XX  GRT
ST  6.70/s   -- -65% -69%
XX  19.1/s 186%   -- -13%
GRT 21.9/s 227%  14%   --
[download]