I'm afraid your benchmark left room for improvement, KM:

use Benchmark;
 
          $string = "hello world I am a string of some sort of length"
+;
 
          timethese(-1, {
              "match" => sub { ($char) = $string =~ /^.{4}(.)/;},
              "unpack" => sub { ($char) = unpack("x4 A1", $string);},
              "substr" => sub { ($char) = substr($string,4,1);},
              "silly"  => sub { my $string2 = $string; $char = chop $s
+tring2 while length $string2 >= 4;}, #note the local
              });
[download]

Which gave somewhat more expected results:

Benchmark: running match, silly, substr, unpack, each for at least 1 C
+PU seconds...
     match:  3 wallclock secs ( 1.00 usr +  0.02 sys =  1.02 CPU) @ 18
+0382.35/s (n=183990)
     silly:  4 wallclock secs ( 1.15 usr +  0.09 sys =  1.24 CPU) @ 17
+341.13/s (n=21503)
    substr:  2 wallclock secs ( 1.00 usr +  0.05 sys =  1.05 CPU) @ 94
+4029.52/s (n=991231)
    unpack:  2 wallclock secs ( 1.02 usr +  0.06 sys =  1.08 CPU) @ 36
+2192.59/s (n=391168)
[download]

Nevertheless, I'm suprised at the speed of silly. Can anybody explain why eating that string is faster than unpack, even if the string has some length?

Cheers,

jeroen
"We are not alone"(FZ)
Update: Ouch! Fixed... silly is the slowest, and the universe is back to normal :-)
The last benchmarks were recorded when my CPU hit a load of 4.63, so they may be a little distorted ;-)