I think it would make things easier to have a single HoH keyed by your "flag" values with sub-hashes of "contig" and count key/value pairs. This would facilitate accessing everything via keys rather than having to pass separate hashes into a subroutine. It might also look a bit neater if you use printf to make columns line up. Something along these lines perhaps?

knoppix@Microknoppix:~$ perl -Mstrict -Mwarnings -E '
> my %counts = (
>    Bact      => { A => 3, B => 7, C => 6         },
>    Euk       => {         B => 5, C => 9         },
>    Proph_Vir => { A => 8,         C => 4, D => 2 },
>    );
> my @sortedKeys = do {
>    my %seen;
>    sort grep { not $seen{ $_ } ++ }
>       map { keys %{ $counts{ $_ } } }
>       keys %counts;
>    };
> 
> printf qq{%-10s%4s%4s%4s%4s\n}, q{}, @sortedKeys;
> foreach my $flag ( sort keys %counts )
> {
>     printf qq{%-10s%4s%4s%4s%4s\n},
>        $flag,
>        map { exists $counts{ $flag }->{ $_ }
>              ? $counts{ $flag }->{ $_ }
>              : 0
>            } @sortedKeys;
> }'
             A   B   C   D
Bact         3   7   6   0
Euk          0   5   9   0
Proph_Vir    8   0   4   2
knoppix@Microknoppix:~$
[download]

A minor niggle with your code; why chomp comment lines and then discard them? It would make more sense to swap the two lines so you only chomp data lines.

I hope this is helpful.