I would not change any other constructs, nor the way $_ is localised.

I would prefer it if this

my $i;
for $i ( 0 .. 10 ) {
    last if <somecondition>;
    ...
}
my $x = substr $somestring, 0, $i;
[download]

was consistant with

my $i=0;
{
    ...
    last if <somecondition>;
    $i++;
    redo;
}
my $x = substr $somestring, 0, $i;
[download]

or

$i=0; 
do{ 
    ...
    $i++ 
} until <somecondition>;
my $x = substr $somestring, 0, $i;
[download]

If you stop getting hung up on the magic of $_, which I wouldn't change anyway, I think that lexicals that automagically revert to some previous value after the programmer has explicitly modified that value is a mysterious and non-useful behaviour that definitely doesn't DWIM.

If the programmer wants that behaviour, he could still get it by using nested scopes to achieve it

my $i=42; 
for my $i (1..10){ 
    print $i; 
} 
print $i;
## gives
1 2 3 4 5 6 7 8 9 10 42
[download]

As is, the current behaviour is a special case (of a pre-existing lexical that gets automagically localised), that doesn't fit the pattern of other looping constructs (map & grep can't use lexicals as their iterator variables) and a special case that removes flexibility, adds nothing and is confusingly non-useful.