DOM parsing is costly, I have considered that approach, but I would do it only in the design phase of the application and deployment, or after changes made to some template, its to say I wouldn't use it as the page generating core.

I ran benchmark on a 5649 Byte html file with HTML::Seamstress and Servlet.

The only changes were uncomment use warnings because it clutters up the test with warnings about unitialized variables, and changed the warn "PROCESS_TREE: ", $tree->as_HTML; to a print statement.
No replace_content at all.

bench.pl:

#!/usr/bin/perl

use lib ".";
use Benchmark qw(:all) ;
use Servlet qw(servlet);
use example; # the spkg.pl generated package

open(O,">/dev/null"); select O;

# initialize some empty references I need in example.al
my $freelist = [];
my $list = [];
my $in = {};

my $r = example->new;
my $s = servlet(
    in          => 'example.html',
    args        => [ qw($nm $list $em8 $error $in $scopename $ref $res
+ult $em26 $link $freelist $networkname $em14 $em9 %oui) ],
);

my $count = 10000;

my $results = timethese($count, {
    'Seamstress' => sub { $r->process() },
    'Servlet'    => sub { $s->($nm, $list, $em8, $error, $in, $scopena
+me, $ref, $result, $em26, $link, $freelist, $networkname, $em14, $em9
+, %oui) },
});
select STDOUT;
cmpthese($results);
[download]

              Rate Seamstress    Servlet
Seamstress   242/s         --       -99%
Servlet    47619/s     19548%         --
[download]

greetings,
--shmem

_($_=" "x(1<<5)."?\n".q·/)Oo.  G°\        /
                              /\_¯/(q    /
----------------------------  \__(m.====·.(_("always off the crowd"))."·
");sub _{s./.($e="'Itrs `mnsgdq Gdbj O`qkdq")=~y/"-y/#-z/;$e.e && print}