I'd be interested to see your benchmark (code + data), as I don't come to the same conclusion.

Test and output attached below. Looks like it is dependent on your data set...

use strict;
use Benchmark 'cmpthese';

my @data = <DATA>;
my @long = map { join '', $_ x 100 } @data;

my %cases = (
  'Single' => sub { for ( @long ) { /^\w+(?:,\w+)*$/ } },
  'Double' => sub { for ( @long ) { /^[\w,]+$/ && ! /^,|,,|,$/ } },
);

cmpthese( 0, \%cases);

__DATA__
!@#$as3dfa
,sdfas3df,
asd3fsa,,a3sdf
as3df,asdf3,3asdf,asd3f
sad3fasdjasdfkasdfklas3jf
3sad3fasdjasdfkasdfklas3jf
3sad3fasdjasdfkasdfklas3jf3
[download]

          Rate Single Double
Single  6158/s     --   -83%
Double 35319/s   474%     --