So, restricting to solutions that do not compress internal spaces, and ignoring the fact that " " is identical to \s, I am surprised that split is faster than tr. Converted the one-liner to a script for readability):

$ cat test.pl
use 5.18.0;
use warnings;

use Benchmark qw( cmpthese );

my $x = join "    " => "", ("abc") x 5, "";

sub splt {
    split " ", reverse ((split " ", (reverse $x), 1)[0]), 1;
    }
sub rgx {
    $x =~ s/^\s+//r =~ s/\s+$//r;
    }
sub trx {
    my $y = $x =~ y/ /x/cr;
    substr ($x, index ($y, "x"), rindex ($y, "x") - length ($x) + 1);
    }

say "sourc: |$x|";
say "split: |", splt (), "|";
say "regex: |", rgx  (), "|";
say "tr/x/: |", trx  (), "|";

cmpthese (-2, { splt => \&splt, rgx => \&rgx, trx => \&trx });
[download]

$ perl test.pl
sourc: |    abc    abc    abc    abc    abc    |
split: |abc    abc    abc    abc    abc|
regex: |abc    abc    abc    abc    abc|
tr/x/: |abc    abc    abc    abc    abc|
          Rate  rgx  trx splt
rgx  1047602/s   -- -59% -65%
trx  2553722/s 144%   -- -14%
splt 2958598/s 182%  16%   --
[download]

So far, none of the presented alternatives to s{^\s+}{}r and friends appeal to me, even though twice as fast or more, I would definitely choose the regex one over the magic of the other two. YMMV.

Here's the one that also squeezes internal spaces with no \s, and there I would seriously consider the join/split variant:

$ cat test.pl
use 5.18.0;
use warnings;

use Benchmark qw( cmpthese );

my $x = join "    " => "", ("abc") x 5, "";

sub splt {
    join " " => split " " => $x;
    }
sub rgx {
    $x =~ tr/ / /sr =~ s/^ //r =~ s/ $//r;
    }
sub trx {
    my $y = $x =~ tr/ /x/cr;
    substr ($x, index ($y, "x"), rindex ($y, "x") - length ($x) + 1) =
+~ tr/ / /sr;
    }

say "sourc: |$x|";
say "split: |", splt (), "|";
say "regex: |", rgx  (), "|";
say "tr/x/: |", trx  (), "|";

cmpthese (-2, { splt => \&splt, rgx => \&rgx, trx => \&trx });
[download]

$ perl test.pl
sourc: |    abc    abc    abc    abc    abc    |
split: |abc abc abc abc abc|
regex: |abc abc abc abc abc|
tr/x/: |abc abc abc abc abc|
          Rate  rgx splt  trx
rgx  1012851/s   -- -42% -52%
splt 1739341/s  72%   -- -17%
trx  2105282/s 108%  21%   --
[download]

substr($x, index($_ = $x =~ tr/ /x/cr, 'x'), rindex($_, 'x') - length(
+$x) + 1)
[download]

Does not matter what variable you use:

# perl -e ' { $line="   aaa bbb ccc ddd eee fff    ";  $line=substr($l
+ine,index($_=$line=~tr/ /x/cr,'x'),rindex($_,'x')-length($line)+1); p
+rint qq(|$line|\n)}'
|aaa bbb ccc ddd eee fff|
[download]

Sorry, I used a different variable name. The real difference was the placement of the parentheses.
map{substr$_->[0],$_->[1]||0,1}[\*||{},3],[[]],[ref qr-1,-,-1],[{}],[sub{}^*ARGV,3]