I can think of two reasons to not use die/eval for (normal) control flow:

(1) Clarity. An exception is supposed to indicate — well, an exceptional condition: either an error, or a resource failure. Throwing an exception as a part of normal execution is liable to mislead maintainers of the code when they try to understand what’s going on. So, at the very least, it’s poor style.

(2) Efficiency. Consider the following code:

#! perl
use strict;
use warnings;
use Benchmark qw(cmpthese);

cmpthese(1_000_000, {
    'bare_loop' => \&bare_loop,
    'eval_loop' => \&eval_loop,
});

sub bare_loop
{
    my ($count, $evens, $flag) = (100, 0, 0);

    while ($count--)
    {
        next if $count % 2;     # odd

        if ($count == 50)
        {
            $flag = 1;
            last;
        }

        ++$evens;
    }

    print "Normal loop exit\n" unless $flag;
}

sub eval_loop
{
    my ($count, $evens) = (100, 0);

    eval
    {
        while ($count--)
        {
            next if $count % 2;     # odd
            die  if $count == 50;
            ++$evens;
        }
    };

    print "Normal loop exit\n" unless $@;
}
[download]

Typical output (on my machine):

             Rate eval_loop bare_loop
eval_loop 71541/s        --      -23%
bare_loop 92362/s       29%        --
[download]

So, there is a definite performance penalty for throwing and then catching the exception. Not nearly as great a penalty as in a language like C++, but still — why opt for a less efficient method when the more efficient methods are just as easy to use?