what can we gain from changing this? except the bc break?

Sheer logic.
What remains from a 4-character string after cutting 4 characters? An empty 
string.

In practice it would allow for a cleaner code, like:
---------
public static function endsWith($string, $suffix) {
		$suffixLength = strlen($suffix);
		return $suffix === substr($string, -$suffixLength);
	}
--------
Method endsWith returns true for:
endsWith("kebab", "ebab");
endsWith("kebab", "bab");
endsWith("kebab", "ab");
endsWith("kebab", "b");

but it returns false for
endsWith("kebab", "kebab");

<?
public static function endsWith($string, $suffix) {
  return preg_match("/{$suffix}$/", $string);
}
?>

No need to rewrite the language :)

<?
public static function endsWith($string, $suffix) {
  return 1 === preg_match("/{$suffix}$/", $string);
}

It should return a boolean :)
Thanks :)

------
Still, it would be more logical if substr returned an empty string. I guess 
backward compatiblity is more important than consistency.

From http://tr.php.net/manual/en/function.substr.php: 
If string is less than or equal to start characters long, FALSE will be 
returned.
would become:
If string is less than start characters long, FALSE will be returned.

Unless there is some specific compatibility issue that would be raised if your change were implemented, I agree.