Have you considered something like this for your application?:

#! perl -slw
use strict;

sub bestN {
    my( $cmp, $n, $ref ) = @_;
    my $src = $ref;
    if( ref $ref eq 'ARRAY' ) {
        my $t = 0;
        $src = sub{ $t < @{ $ref } ? $ref->[ $t++ ] : undef };
    }
    my $_cmp = sub {
        local( $::a, $::b ) = @_;
        $cmp->()
    };
    my @top = sort $cmp map $src->(), 1 .. $n;
    while( defined( $_ = $src->() ) ) {
        next if $_cmp->( $top[ $#top ], $_ ) <= 0;
        my $p = $#top;
        while( $p > 0 ) {
            $_cmp->( $top[ --$p ], $_ ) <= 0 and last;
        }
        splice @top, $p, 0, $_;
        pop @top;
    }
    return @top;
}

my @in = 1 .. 100;
print for bestN sub{ $::b <=> $::a }, 10, \@in;

open DICT, '<', $ARGV[0] or die $!;
print for bestN sub{ length( $::b ) <=> length( $::a ) }, 10, sub{ loc
+al $^W; chomp( $_ = <DICT> ); $_ };;
[download]

Output:

C:\test>topN words.txt
100
99
98
97
96
95
94
93
92
91
abstractionisms
abstractionists
acanthocephalan
acceptabilities
acceptingnesses
accessibilities
acclimatization
accommodatingly
accommodational
absorbabilities
[download]