I'd be interested in seeing your results. I did some some benchmarking too, and found that the regex is slower.

Here's the program (using my soon to be released 'Benchmark::Sized' module) and the results.

#!/usr/bin/perl
 
use strict;
use warnings qw 'all';

use Benchmark::Sized;

use vars qw /@array/;

my $stats = sized_timethese
    run_for    =>    2, 
    steps      =>    8, 
    start_size =>    1, 
    end_size   => 1000,
    seed       => sub {
        @main::array = map {join "" => map {chr rand 128} 1 .. $_ [0]}
+ 1 .. 20
    },
    code       => {
        regex  => 'my @foo = grep {! /^\.\.?\Z/} @main::array',
        ne     => 'my @foo = grep {$_ ne "." and $_ ne ".."} @main::ar
+ray',
    }
;

print "Runs/second for regex vs ne:\n";
print_timed $stats, type => 'sized';

__END__
Runs/second for regex vs ne:
Size         ne    regex
   1:  20744.28 12809.79   
   3:  22549.51 13048.06
   7:  23110.45 12337.91
  19:  21600.00 12560.28
  52:  21005.50 12276.56
 139:  19269.95 11891.63
 373:  18345.79 11220.29
1000:  14422.54  9634.74
[download]

Abigail

Unfortunately, I don't have your shiny Benchmark::Sized (I'm looking forward to its release), but there are my results using Benchmark::cmpthese:

2;130 juerd@ouranos:~/modje/Blah/Acme/Tie/Handle/Blah/Acme-Tie-Handle-
+Blah--0.10$ perl -MBenchmark=cmpthese -e'$_="fooooo"; cmpthese(-1, {r
+e => sub {$a = !/^\.\.?\z/}, ne => sub { $a = ($_ ne "." and $_ ne ".
+."); }})'
Benchmark: running ne, re, each for at least 1 CPU seconds...
        ne:  1 wallclock secs ( 1.02 usr + -0.01 sys =  1.01 CPU) @ 12
+18808.91/s (n=1230997)
        re:  1 wallclock secs ( 1.09 usr +  0.00 sys =  1.09 CPU) @ 18
+03743.12/s (n=1966080)
        Rate   ne   re
ne 1218809/s   -- -32%
re 1803743/s  48%   --
[download]

Maybe this is because you used \Z, which is equivalent to $ if /s is not used. \z should be used (explained elsewhere in this thread).

2;0 juerd@ouranos:~/modje/Blah/Acme/Tie/Handle/Blah/Acme-Tie-Handle-Bl
+ah--0.10$ perl -MBenchmark=cmpthese -e'$_="foo"; cmpthese(-0.3, { z =
+> sub { /o\z/ }, Z => sub { /o\Z/ } });'
Benchmark: running Z, z, each for at least 0.3 CPU seconds...
         Z:  2 wallclock secs ( 0.43 usr +  0.02 sys =  0.45 CPU) @ 36
+70015.56/s (n=1651507)
         z:  1 wallclock secs ( 0.44 usr + -0.01 sys =  0.43 CPU) @ 44
+19869.77/s (n=1900544)
       Rate    Z    z
Z 3670016/s   -- -17%
z 4419870/s  20%   --
[download]

- Yes, I reinvent wheels.
- Spam: Visit eurotraQ.