Very nice, mirod++, a good explanation! I just add the compact version - which is also a bit more efficient because it uses a hash slice:

my %unique;

@unique{ split /,\s*/, ( split /\|/ )[2] } = () while (<DATA>);

my @unique_names = keys %unique;
print join "-", @unique_names;
print "\n";

__DATA__
value 1|value 2|tom, chris, dave|value 4
value 1|value 2|peter, tom, dave|value 4
[download]

-- Hofmator