Indeed you are correct, my original code is flawed as you describe. Thanks for pointing out my error Cees++

I've re-run a subset of the tests using the following slightly modfied version which corrects this error and the correction slows the insertion time by approximately 1 second for the million keys. The iteration time comes out exactly the same.

print time;
$hash{ substr $_, 0, 4 } .= $/. $_ for 1 .. 1_000_000;
$hash{$_} .= $/ for keys %hash; # Append a final separator
print time;

1069484550
1069484566 # 16 seconds.

$n=0;
print time;
for ( 1 .. 1_000_000 ) {
    # Ensure no partial matches by prepending and appending  
    # separators to the search key.
    next unless exists $hash{ substr $_, 0, 4 }
            and 1+index( $hash{ substr $_, 0, 4 }, $/.$_.$/ );
    $n++;
}
print time, $/, $n;

1069484612
1069484656 # 44 secs.
 1000000
[download]