One problem with your existing code is that you are counting the lines $outsize++;, not the bytes, and comparing that line count against your chunksize if( $outsize>$chunksize ..., which means your 'small' files are going to end up containing 500,000 lines! If your lines average 67 byte per line, then each output file would be around 33MB.

You need to change that to be $outsize += length() + 1; to achieve your original aim.

Update: I noticed the $outsize++; at the top, but not the $outsize += length at the bottom of the loop. Why two statements?

Beyond that, to achieve your second aim, you will need to buffer the contents of each file and accumulate a second per file count of the bytes read, and delay writing until you have accumulated a complete invoice.

Only at that point will you be able to determine whether to write the buffer out to the current composite file, or to a separate file, dependant upon how big it is. Remember to subtract the accumulated per file count from $outsize when you write individual files. Or only accumulate to $outsize when you have written to the composite output file.

It would also save time if you tested for the start-of-invoice condition using substr, eg.

if( substr( $_, 67, 2 ) eq '11' ) {
...
[download]

It gets expensive running up the regex engine on every line of files of this size.

Unless you want special handling for the large invoices, it looks like your code is already doing that. Consider the following mutation of your code for demonstration purposes:

use warnings;
use strict;

my $chunksize = 50;
my $filenumber = 0;
my $outsize = $chunksize + 1;
my $eof = 0;

while (<DATA>) {
    if ($outsize > $chunksize and /^11/) {
        $outsize = 0;
        $filenumber++;
        print "*** New file number $filenumber\n";
    }

    print "$_";
    $outsize += length;
}

__DATA__
11 The first invoice.
We get this whole thing even though it exceeds the 50 character limit
because the conditional code requires an invoice marker before it will
start a new file.
11 Second
11 Third
11 Fourth invoice
11 Fifth invoice
This one is longer, but not excessive
11 Sixth: break before because limit hit in fifth
11 seventh and final
[download]

Prints:

*** New file number 1
11 The first invoice.
We get this whole thing even though it exceeds the 50 character limit
because the conditional code requires an invoice marker before it will
start a new file.
*** New file number 2
11 Second
11 Third
11 Fourth invoice
11 Fifth invoice
This one is longer, but not excessive
*** New file number 3
11 Sixth: break before because limit hit in fifth
11 seventh and final
[download]

How would the output be different to achieve what you want?