Slightly correcting: It won't use Latin-1 but just do nothing. I tested with a Unicode character (≈ "almost equal") and it returned the character readable.
Can be tested by passing html_entity_decode to htmlspecialchars and then echo'ing.

The very issue was already addressed and the appropriate fix is ready for php5, though we won't introduce this feature to the current stable version (4.3.x).

See:
http://cvs.php.net/diff.php/php-src/NEWS?r1=1.1403&r2=1.1404

OK, I found another way for my issue anyway...

<?php
// Returns the utf string corresponding to the unicode value (from php.net, courtesy - romans@void.lv)
function code2utf($num)
{
	if ($num < 128) return chr($num);
	if ($num < 2048) return chr(($num >> 6) + 192) . chr(($num & 63) + 128);
	if ($num < 65536) return chr(($num >> 12) + 224) . chr((($num >> 6) & 63) + 128) . chr(($num & 63) + 128);
	if ($num < 2097152) return chr(($num >> 18) + 240) . chr((($num >> 12) & 63) + 128) . chr((($num >> 6) & 63) + 128) . chr(($num & 63) + 128);
	return '';
}

function encode($str)
{
	return preg_replace('/&#(\\d+);/e', 'code2utf($1)', utf8_encode($str));
}
?>

Is there a particular reason you won't backport this fix to the 4.3 series? I can't find an explanation/discussion of this refusal. If somebody else backported this fix, would it be accepted? Sounds like quite a fairly important and useful bugfix to me.

Just curious. Currently downloading PHP5, will give it a spin.