I think I like the Text::CSV or Text::ParseWords module-based parsing approaches a bit better, but if you have Perl version 5.10+, you can use a regex approach that's almost identical to your OPed regex if you take advantage of the  (?|...|...) "branch reset" construct (see Extended Patterns in perlre):

c:\@Work\Perl\monks>perl -wMstrict -le
"use 5.010;
 ;;
 use Data::Dump qw(dd);
 ;;
 my $s = '\"word1\",word2,\"word3,word4\"';
 ;;
 my @ra = $s =~ m{ (?| \"([^^\x22]*)\" | ([^,]+) ) (?:,|$) }xmsg;
 dd \@ra;
"
["word1", "word2", "word3,word4"]
[download]

(Note that what appears above as  \"([^^\x22]*)\" due to the peculiarities of Windose command line interpretation and the limitations of my personal REPL should really be  "([^\x22]*)" or better yet  "([^"]*)")