how clustered are your lines? here are a couple tweakable previous line recognizers.

#
# remember everything, probably uses too much memory.
#
{
    my %seen;
    sub seen_complete {
        return 1 if exists $seen{$_[0]};
        $seen{$_[0]} = ();
        return 0;
    }
}

#
# remember last N lines only.
#
{
    my %seen;
    my $remember = 200;
    my @memory;
    sub seen_fixed {
        return 1 if exists $seen{$_[0]};
        delete $seen{shift(@memory)} if @memory > $remember;
        push @memory, $_[0];
        $seen{$_[0]} = ();
        return 0;
    }
}

#
# remember N buckets of lines with X lines per bucket.
#
{
    my @bucket = ( {} );
    my $numbuckets = 2;
    my $bucketsize = 200;
    sub seen_bucket {
        foreach (@bucket) {
            return 1 if exists $_->{$_[0]};
        }
        if (keys %{$bucket[-1]} >= $bucketsize) {
            shift @bucket if @bucket >= $numbuckets;
            push @bucket, {};
        }
        $bucket[-1]->{$_[0]} = ();
        return 0;
    }
}
[download]

i only tested the last one, and only sorta tested it at that.

while (<>) {
    print unless seen_bucket($_);
}
__END__

Ten sets of 1..400 should get uniq'd to 1..400

$ perl -le 'for(1..10){for(1..400){print}}' | perl dup.pl | wc -l
    400

Ten sets of 1..401 should get uniq'd to (1..401) x 10 because
2 buckets of 200 lines hold upto 400 

$ perl -le 'for(1..10){for(1..401){print}}' | perl dup.pl | wc -l
   4010
[download]