I point out that solutions which perform set unions pair-wise, iteratively on the set of lists do not scale well. Consider the worst-case scenario, where all of the lists are identical, and there's a million of them.

A better approach is the counting one, like that suggested by dtr, but uniqifying each list by converting it to its corresponding set hash, as basje did.

my %counts;

for my $ar ( @arrays )
{
    my %set;
    @set{ @$ar } = (); # this is faster.

    $counts{$_}++ for keys %set;
}

my @union = grep { $counts{$_} == @arrays } keys %counts;
[download]