This approach is a variation on AnonyMonk's above:

c:\@Work\Perl>perl -wMstrict -le
"use 5.010;
 ;;
 my $s = qq{Please;, read these&\x23xB0;, before you post 180;, if not
+ absolutely &\x23x2013;, sure.};
 print qq{'$s'};
 ;;
 my $entity = qr{ & [\x23] x [[:xdigit:]]+ ; }xms;
 ;;
 $s =~ s{ (?: $entity , (*SKIP) (*FAIL))?  ; \K , }{}xmsg;
 print qq{'$s'};
"
'Please;, read these°, before you post 180;, if not absolutely &#
+x2013;, sure.'
'Please; read these°, before you post 180; if not absolutely &#x2
+013;, sure.'
[download]

Some notes (update: now with more numbers for your indexing enjoyment):

1. My REPL does not like the # character at all, so it has to be represented as  \x23 in strings and regexes, but it prints and otherwise operates as expected;
2. The substitution regex uses features added with Perl version 5.10:  (*SKIP) (*FAIL) (see Special Backtracking Control Verbs), and  \K (see Extended Patterns);
3. The  $entity regex is very naive (as is the one used by AnonyMonk) and you will have to come up with one that covers all your cases;
4. The  (?: $entity , (*SKIP) (*FAIL))? sub-pattern acts as a variable width negative look-behind by 'consuming' (i.e., skipping past) something that is needed for a subsequent match, in this case a comma following an entity.

This worked well for me. I learned something new (*SKIP) (*FAIL). Thank you

s/&#x\w+;,\K|;\K,//g
[download]

This worked well for me. Thank you.

How about no look behind?

match 
entity # $1
commatarget # $2
and replace entity with itself and commatarget with comma, and you're 
+done :)
[download]

As soon as you try making it any valid entity format, you're back to no look behind because its fixed width only http://perldoc.perl.org/perlre.html#%28?%3C!pattern%29

my $f = 'blah;, ente°, blah;, blah.';
$f =~ s{(?<!&[#]x[A-Za-z0-9]{2});,}{,}g;
print "$f\n";
__END__
blah, ente&#xB0;, blah, blah.
[download]