This uses threads to achieve a near perfect 75% reduction on overall runtime using 4 threads:

#! perl -slw
use strict;
use Time::HiRes qw[ time ];
use threads;
use threads::shared;

my @buffers;

sub thread {
    my $tid = threads->tid;
    print "[$tid] starting";
    my $bufRef = shift;
    open my $fh, '<', $bufRef;
    while( <$fh> ) {
        ## next if ... ## add pre-filter criteria here is applicable
        my @f = split "\t", $_;
        ## do something with the data
    }
    print "[$tid] ending";
}

our $N //= 199;
our $T //= 4;

my $start = time;
my @threads;
for( 0 .. $N-1 ) {  ## glob '*.tsv or whatever
    my $bufRef = \$buffers[ $_ % $T ];
    {
        local( @ARGV, $/ ) = 'numbers.tsv';
        $$bufRef = <>;
    }
    push @threads, async( \&thread, $bufRef );

    while( @threads >= $T ) {
        sleep 1 until threads->list( threads::joinable );
        for( reverse 0 .. $#threads ) {
            next unless $threads[ $_ ]->is_joinable;
            $threads[ $_ ]->join;
            splice @threads, $_, 1;
        }
    }
}
$_->join for @threads;

printf "Took %f seconds\n", time() - $start;

__END__
C:\test>1036737-3 -T=1 -N=8
[1] starting
[1] ending
[2] starting
[2] ending
[3] starting
[3] ending
[4] starting
[4] ending
[5] starting
[5] ending
[6] starting
[6] ending
[7] starting
[7] ending
[8] starting
[8] ending
Took 160.641885 seconds

C:\test>1036737-3 -T=4 -N=8
[1] starting
[2] starting
[3] starting
[4] starting
[2] ending
[1] ending
[5] starting
[4] ending
[3] ending
[6] starting
[7] starting
[8] starting
[5] ending
[6] ending
[7] ending
[8] ending
Took 39.710515 seconds
[download]