Because capturing is the most expensive thing that happens there. On a ten minute program I shaved a minute off by avoiding capturing groups (you mistook (?>...) for a capture block) and using substr instead. A minute is pretty small for the thirty megabyte data set I was working with, I just expected most data sets to be half a gig or more. I picked a simpler example for this code sample - it still illustrates the point (and verified through dprof).

unless ($$buffer =~ m/^H=(?>\d+);/) {
    confess "$rec_num was missing the header prefix m/^H=\\d+;/: $$buf
+fer";
}

my $header_length = substr $$buffer, 2, $+[0]-3;

# vs

unless ($$buffer =~ m/^H=((?>\d+));/) {
    confess "$rec_num was missing the header prefix m/^H=(\\d+);/: $$b
+uffer";
}

my $header_length = $1;
[download]