为什么S/^\S+\S+$//G;比两个单独的替换慢得多?

~# debugperl -Dr -e 's/^\s+//g'
Compiling REx `^\s+'
size 4 Got 36 bytes for offset annotations.
first at 2
synthetic stclass "ANYOF[\11\12\14\15 {unicode_all}]".
   1: BOL(2)
   2: PLUS(4)
   3:   SPACE(0)
   4: END(0)
stclass "ANYOF[\11\12\14\15 {unicode_all}]" anchored(BOL) minlen 1

# debugperl -Dr -e 's/^\s+|\s+$//g'
Compiling REx `^\s+|\s+$'
size 9 Got 76 bytes for offset annotations.

   1: BRANCH(5)
   2:   BOL(3)
   3:   PLUS(9)
   4:     SPACE(0)
   5: BRANCH(9)
   6:   PLUS(8)
   7:     SPACE(0)
   8:   EOL(9)
   9: END(0)
minlen 1 

use strict;
use warnings;
use Benchmark qw(cmpthese);

my $ws = "   \t\t\n";

for my $sz (1, 10, 100, 1000){
    my $str = $ws . ('Z' x $sz) . $ws;
    cmpthese(-2, {
        "alt_$sz" => sub { $_ = $str; s/^\s+|\s+$//g },
        "sep_$sz" => sub { $_ = $str; s/^\s+//; s/\s+$// },
    });
}

           Rate alt_1 sep_1
alt_1  870578/s    --  -16%
sep_1 1032017/s   19%    --

            Rate alt_10 sep_10
alt_10  384391/s     --   -62%
sep_10 1010017/s   163%     --

            Rate alt_100 sep_100
alt_100  61179/s      --    -92%
sep_100 806840/s   1219%      --

             Rate alt_1000 sep_1000
alt_1000   6612/s       --     -97%
sep_1000 261102/s    3849%       --

^\s+ \s+$