Show your benchmark. I see about a 75x difference between the single regex and the multiple regex solutions offered by the OP.

Note that benchmarks are rather like statistics: Lies, damn lies and benchmarks.

I have a feeling I may be about to learn something... ;-)

use strict;
use warnings;


use Benchmark   qw( cmpthese );

my $results = cmpthese( -10, {
                              'r3' => sub {
                                  my $string = "  stuff  ";
                                  $string =~ s/^\s//g;
                                  $string =~ s/\s$//g;
                                  $string =~ s/\s+$//g;
                              },
                              'r1' => sub {
                                  my $string = "  stuff  ";
                                  $string =~ s/^\s|\s$|\s+$//g;
                              },
                             }
                      );
[download]

---

       Rate  r1  r3
r1 295926/s  -- -1%
r3 299980/s  1%  --
[download]

---

Updated: Now that I see GrandFather's detailed Benchmark below, I see that my error was to use a short string. When I change to

my $string = (' ' x 1000) . 'x' . (' ' x 1000);
[download]

matching GrandFather's, I get the following:

       Rate   r3   r1
r3  61890/s   -- -79%
r1 297607/s 381%   --
[download]

So Lies, damn lies, and benchmarks (with the wrong data), indeed.

Changing "  stuff  " to (' ' x 1000) . 'x' . (' ' x 1000) gives:

       Rate    r1    r3
r1   1849/s    --  -99%
r3 166166/s 8888%    --
[download]

It is important that a benchmark test what you think it is testing so what you think is being tested must entail a significant portion of the benchmark's processing time.

Update: BTW, my result for your original benchmark was:

       Rate   r1   r3
r1 767220/s   -- -10%
r3 855450/s  11%   --
[download]

---

Perl reduces RSI - it saves typing