Hi,

if the text body can be split flatly (that is not nested)
into parts which need mangling
and into parts to be left alone
it would be easy.
Then, you could loop through the text like this
(but maybe you should not use the patterns from below:-)

use strict;
use warnings;

my $text = join q{}, <DATA>;
my $newtext = q{};

while ($text =~ m{
  \G              # start where we left 
  (.*?)           # text to be mangled
  (?:             # followed by
    STARTPATTERN  # the start of verbatim mark
    (.*?)         # the text to be taken verbatim
    ENDPATTERN    # the end of verbatim mark
  )?              # must not be always present
}xmsg) {

  # add the mangled version of the text
  $newtext .= mangle($1);

  if (defined $2)
  {
    # if there is a verbatim part, copy it verbatim
    $newtext .= $2;
  }
}
print $newtext;
exit 0;

sub mangle
{
    return uc $_[0];
}
__DATA__

Once upon STARTPATTERNa timeENDPATTERN there 
was a Perl monk. STARTPATTERNBy steadily trainingENDPATTERN 
the great virtues he became very wise...
[download]

gives

ONCE UPON a time THERE 
WAS A PERL MONK. By steadily training 
THE GREAT VIRTUES HE BECAME VERY WISE...
[download]

Hope this helps