Can you tell my why physi's code is slower than the rest? He suggested this:

$data =~ s/(^\s*|\s*$)//g;

I added these two subs to your benchmark:

sub both_at_once {
    my $data = $testdata;
    $data =~ s/(^\s+|\s+$)//g;
    return $data;
}

sub both_at_once2 {
    my $data = $testdata;
    $data =~ s/(^\s*|\s*$)//g;
    return $data;
}
[download]

And this was the result:

Benchmark: timing 100000 iterations of both_at_once, both_at_once2, do
+tstar, first_n_last_1, first_n_last_2...
both_at_once: 10 wallclock secs ( 9.04 usr +  0.00 sys =  9.04 CPU) @ 
+11061.95/s (n=100000)
both_at_once2: 11 wallclock secs (10.40 usr +  0.00 sys = 10.40 CPU) @
+ 9615.38/s (n=100000)
   dotstar:  9 wallclock secs ( 8.30 usr +  0.00 sys =  8.30 CPU) @ 12
+048.19/s (n=100000)
first_n_last_1:  6 wallclock secs ( 5.77 usr +  0.00 sys =  5.77 CPU) 
+@ 17331.02/s (n=100000)
first_n_last_2:  2 wallclock secs ( 2.31 usr +  0.00 sys =  2.31 CPU) 
+@ 43290.04/s (n=100000)
[download]

Unless I'm mistaken, the pattern alternation (^\s+|\s+$) will try to match both patterns on every character. But, does the engine not know to disregard the ^\s+ except at the beginning of the string, and likewise for \s+$, only trying to match at the end? Just curious as to why this is so slow.