I looked at a small number of variations around the second case here:

$x = "foo\x{100}";
chop($y = $x);
$z = substr($x, 0, 1);  Dump($z);  # UTF8
$a = substr($y, 0, 1);  Dump($a);  # not UTF8
[download]

Looking at the perl source, it looks like it treats it differently (and ends up not flagging as UTF8) if the byte length and the character length of the whole source string are the same. (Which is a potential efficiency concern: finding the character length of a large UTF8-flagged string is expensive.)