use strict;
use warnings;
use Benchmark;

my @strings = qw(exception:tex exception:mex asdf);

Benchmark::cmpthese( -5, {
    'one' => sub { my @filtered = grep { /exception:(?!tex)/} @strings
+; },
    'two' => sub { my @filtered = grep { /exception/ && !/tex/ } @stri
+ngs; },
});
__END__
        Rate  one  two
one 175458/s   -- -15%
two 207611/s  18%   --
[download]

update: The filter expressions are different. The second one does not care whether tex is before of after exception and doesn't require a ':' after 'exception'. I have stumbled on a patch of my ignorance trying to make the second match the same strings as the first. The best I have so far is:

use strict;
use warnings;
use Benchmark;

my @strings = qw(exception:tex exception:mex asdf tex:exception:mex);

Benchmark::cmpthese( -5, {
    'one' => sub { my @filtered = grep { /exception:(?!tex)/} @strings
+; },
    'two' => sub { my @filtered = grep { /exception/ && !/tex/ } @stri
+ngs; },
    'three' => sub { my @filtered = grep { pos = 0; /exception:/g && !
+/\Gtex/ } @strings; },
});
__END__
          Rate three   one   two
three  99554/s    --  -18%  -38%
one   121692/s   22%    --  -24%
two   161067/s   62%   32%    --
[download]

update: see also strange behavior of grep with global match [resolved]. Adding "pos = 0" is necessary to clear pos between iterations but it would not be necessary if there were only a single iteration. This makes it difficult to compare the efficiency of "/exception/g && !/\Gtex/".