It's even faster if you avoid the sort block, using GRT:

#!/usr/bin/perl

use strict;
use warnings;

use Benchmark qw /timethese cmpthese/;

our @advocates = map {{fld_title => 'Rec' . $_,
                       fld_type  => [qw /National State Local/] -> [ra
+nd 3]}}
                     1 .. shift || 1000;


our %order   = (LOCAL => 'A', STATE => 'B', NATIONAL => 'C');
our %order_r = reverse %order;

our (@limbic, @GRT);

cmpthese -1  => {
    limbic   => '@limbic = sort {$order {uc $a -> {fld_type}} cmp
                                 $order {uc $b -> {fld_type}} or
                                 $a -> {fld_title} cmp $b -> {fld_titl
+e}}
                           @advocates',
    GRT      => '@GRT = map {{fld_title => substr ($_, 1),
                              fld_type  => $order_r {substr $_, 0, 1}}
+}
                        sort
                        map {join "", $order {uc $_ -> {fld_type}},
                                                 $_ -> {fld_title}} @a
+dvocates',
};
                       
                     
my $limbic = join " " => map {$_ -> {fld_title}} @limbic;
my $GRT    = join " " => map {$_ -> {fld_title}} @GRT;

die "Unequal\n" unless $limbic eq $GRT;
__END__
         Rate limbic    GRT
limbic 61.9/s     --   -49%
GRT     121/s    95%     --
[download]

It does lose the case of fld_type, but that's easily fixable.

Abigail