The problem with that -- as coded -- is that it will require 7 concurrent lists, each the same size as the input array.

Assuming an average 128-byte line length, that means it will need to have room for 175 million items on the stack, for a file that contains 25 million records?

The GRT -- code in the same @out = map() sort map() @in; way would only require 5 times as many concurrent items.

And if the GRT is coded in three steps:

1. Prefix added as the array is built:

   my @data; $#data = 25e6; my $i = 0;
   $data[ $i++ ] = pack "NNA*", m[(\d+)\D+(\d+)], $_ while <>;
   [download]

2. sort done in-place:

   @data = sort @data;
   [download]

3. Prefix trimmed in-place:

   substr $_, 0, 8, '' for @data;
   [download]

It requires (for this example of two numeric keys) just 8 bytes/record extra memory (~6%).