doran,
I think you might find the following a bit more efficient and a bit more readable:

# An inline loop is more efficient then a block - thought almost imper
+ceptable
$KP->{ $_->{k} }{ $_->{v} }++ while $kills_sth->fetchrow_hashref();

# Before I did not know if it was safe to assume no negative numbers -
+ now I can
my %top = map { $_ => {'score' => 0, 'pid' => '', 'tid' => ''} } 1 .. 
+10; 

my @items = qw(score pid tid);

for my $pid ( keys %{ $KP } ) {
    SCORE:
    for my $tid ( keys %{ $KP->{$pid} } ) {
        my $score  = $KP->{$pid}{$tid};
        next if $score < $top{10}{score};    # May or may not improve 
+efficiency
        for my $rank ( 1 .. 10 ) {
            if ( $score > $top{$rank}{score} ) {
                (@{$top{$rank}}{@items}, @{$top{$rank + 1}}{@items}) =
+ ($score, $pid, $tid, @{$top{$rank}}{@items});
                next SCORE;
            }
        }
    }
}

for my $rank ( 1 .. 10 ) {
    printf("%.2d. f: %s\tp: %s\tt: %s\n", $rank, @top{@items});
}
[download]

This would be much more efficient if it could be done with SQL. You would have to provide more info to determine that.

Cheers - L~R

Update: 2004-03-17 This is broken. To fix it you would make %top an AoA, handle fence post issues, and splice entries in the middle while popping off the end. This version will randomly lose values. See this for a working example.