In your FIDDLE and your SUB-REF code you seem to be building your call list dynamically. This is a waste.

You assume you are cpu bound. You're right. But you are bound to be, cause you are waiting to read everything, bound to process it all, then waiting to write it. You may be solving the wrong problem.

It is not clear whether you have enough memory to avoid swapping. It is unknown whether you have other tasks running in your environment. Also unknown is whether you have multiprocessors.

Some ideas:

• Do items one by one.
• Do items in batches small enough to fit memory.
• Additionally have multiple programs doing this job, 1 to 3 more processes than cpus, to start.
• Eliminate routine calls, i.e. apple_wash_core_pulp.
• Use a hash to look up your processes:

  # off-hand untried code
  
  PROGRAM_INIT:
  my %proc_table = ( 
     type => \&process,
     apple => \&apple_wash_core_pulp,
     banana => \&banana_bend_hang,
  );
  
  foreach ( @item){
     &$proc_table{$type} if exists $proc_table{$type};
     next;
     unknown_type();
  }
  [download]

• Use an array and constants instead of a hash. Similar to the above but use &$proc_array[$type]. Also your data could loaded into an array of arrays.