Nice work! On my box, it was about 30% faster, and this subtle tweaking is about 25% faster, yet:

sub improved {
    my $lstr = lc shift;
    my ($p, $ntst) = (length($lstr)-1);
    chop($ntst = substr($lstr, $p, 2)) lt $ntst and return 0
        while ($p--);
    1;
}
[download]

while this very similar model is actually slower

sub hobbled {
    my $str =  lc shift;
    my ($p, $x) = (length($str)-1);
    chop($x) lt $x and return 0
        while $x = substr($str, $p--, 2);
    1;
}
[download]