From what I've found in benchmarks, a positive look-ahead assertion:

$match =~ /($start .*?) (?=$start)/xs;
[download]

is just a HAIR slower than a negative look-ahead assertion:

$match =~ /($start (?: (?! $start ) .)*)/xs;
[download]

The first one matches as little from $start as it can until it's followed by $start again. That would actually FAIL for the last record. I'd suggest changing it to

$match =~ /($start .*?) (?= $start | \Z)/xs;
[download]

Where \Z is the absolute-end-of-string anchor. The second one matches from $start as much as possible that isn't $start again.

And I'd like to set the record straight on /s and /m (even though perlre does a good job). The /s modifier ONLY means that the . regex character matches EVERY character (including \n, which it usually doesn't). The /m modifier means that ^ matches the real beginning of a string, or immediately after a \n, and that $ matches immediately before a newline, or at the real end of a string. Normally, the ^ anchor only matches at the REAL beginning of a string, and the $ anchor only matches the REAL end of a string, or immediately before a newline at the REAL end of a string. When using /m, the \A anchor matches the REAL beginning of the string, and the \Z anchor matches the REAL end of the string.