Double zipping xlsx files is not necessary. The following override is the minimum needed for the demonstration above. The _prepare_sst_string_data may be omitted when calling $wb->set_optimization().

{  no warnings 'redefine';

   sub Excel::Writer::XLSX::Workbook::_store_workbook {

      my $self     = shift;
      my $tempdir  = File::Temp->newdir( DIR => $self->{_tempdir} );
      my $packager = Excel::Writer::XLSX::Package::Packager->new();

      # Convert the SST strings data structure.
      $self->_prepare_sst_string_data();

      # Package the workbook.
      $packager->_add_workbook( $self );
      $packager->_set_package_dir( $tempdir );
      $packager->_create_package();

      # Free up the Packager object.
      $packager = undef;
   }
}
[download]

The time for 60 million cells reduced by 16.5 seconds. Thus, reaching 232k cells per second (2.5x for multi-core).

$ time perl write_60mil_parallel.pl
AMS  ---- DONE.
EMEA ---- DONE.
APJ  ---- DONE.
Node List is Done.

real   4m18.715s
user  11m15.291s
sys    0m 3.766s
[download]

For comparison, the time is shy of 11 minutes for 1 core at 92k cells per second.

$ time perl write_60mil_serial.pl
AMS  ---- DONE.
APJ  ---- DONE.
EMEA ---- DONE.
Node List is Done.

real  10m54.015s
user  10m51.045s
sys    0m 3.224s
[download]

Kind regards, Mario