In such a case I would use CDATA sections. First wrap the content of the dodgy elements in <[CDATA[ ... ]]>, then you can parse them without problem:

#!/bin/perl -w

use strict;
use XML::Twig;
use Data::Dumper;

# generate a file where the content of rec1 and something is
# stuck in CDATA sections
my $tmp="tmp";
open( TMP, ">$tmp") or die "$0 cannot open $tmp: $!"; 
while( <DATA>)
  { s{<(rec1|something)>}{<$1><![CDATA[}g;
    s{</(rec1|something)>}{]]></$1>}g;
    print TMP $_;
  }
close TMP;

# sorry, I could not help but use XML::Twig for this
my %data;

my $t= XML::Twig->new( 
         twig_handlers => { r => sub { $data{$_->field( 'key')}= 
                                        { rec1      => $_->field( 'rec
+1'), 
                                          something => $_->field( 'som
+ething')
                                        };
                                       $_[0]->purge; # I like to save 
+memory 
                                      }
                           },
                      );
$t->parsefile( $tmp);
print Dumper( %data);


__DATA__
<data>
  <r>
    <key>k1</key>
    <rec1>data</rec1>
    <something>else</something>
  </r>
  <r>
    <key>k2</key>
    <rec1>includes <br> and non UTF-8 chars like é, or nasties like < 
+</rec1>
    <something>else, <p>ugly <i>too<b>isn't</i> it</b> & all</p></some
+thing>
  </r>
</data>
[download]

The only restrictions are: if the "embedded" data contains ]]> then you must break the CDATA section, and the data of course it must not contain </rec1> or </something>.