Why do overloaded objects bechave differently depending on their construction?

overload-stringify.pl

package Acme::Object::Overloaded;

use overload '""' => sub{ 'String' }, '0+' => sub{ 123 }, 'fallback' => '0+';

sub new{
	bless {}, shift;
}
1;

package main;

use Test::More;
sub new_object { Acme::Object::Overloaded->new(@_) };
is ((new_object() + 1), 124); # passes
is ((new_object + 1), 124); # fails, got: 'String'
is ((new_object() + 1), 124); # passes
done_testing();

# SOLVED by huf: 
# They don't behave differently, the above code contains a subtle bug, as demonstrated below:
#
# > 17:34 < huf> deparse: sub foo {} foo + 1
# > 17:34 < perlbot> huf:  foo(1);
#
# i.e. test 2 is being parsed as `is(new_object(+1), 124);`