The most efficient way to pull data from a database, as per the DBI documentation, is a combination of bind_columns() and fetch(). I'm not quite sure what you're trying to do with undef, but I don't think it will help you any.

If what you want is to pull the set of data from the database, it would be slightly (by not much) faster to do this:

# Error handling assumed.

my $sql = " ... ";
my $sth = $dbh->prepare_cached( $sql );
$sth->bind_columns( \(my ($foo)) );

my @data;
while ($sth->fetch) {
    push @data, $foo;
}
$sth->finish;
[download]

The reason is that Perl doesn't have to allocate and deallocate return values - the data is placed right into the scalar reference you passed bind_columns(). A similar method can be used for your HoA in the third snippet. You can, alternatively, do:

# Error handling assumed.

my $sql = " ... ";
my $sth = $dbh->prepare_cached( $sql );

my @stuff;
$sth->bind_columns( \(my ($foo), @stuff[0..3]) );

my %data;
while ($sth->fetch) {
    $data{$foo} = [ @stuff ];
}
$sth->finish;
[download]

Now, the alternatives I'm presenting won't double your throughput. We're talking, at best, maybe 5-10%, if that. Most applications will see a 0-2% improvement. The guarantee is that this isn't slower and it might be faster.