I see. If you have tags already in your data and want to skip over them then the best plan is to use a proper HTML parser.

If you cannot (or will not) use a parser then a crude plan B which works for your supplied dataset is:

#!/usr/bin/env perl 
use strict;
use warnings;
use Test::More;

my @set = (
    {
        in => 'ABCDEFGHI',
        want => "ABCD\nEFGH\nI\n"
    },
    {
        in => '<b>ABC</b>DEFGHI',
        want => "<b>ABC</b>D\nEFGH\nI\n"
    },
);
plan tests => scalar @set;
my $len = 4;

for my $x (@set) {
    my $i = 0;
    my $out = '';
    my $intag = 0;
    for my $c (split (//, $x->{in})) {
        $out .= $c;
        $intag++ if $c eq '<';
        $intag-- if $c eq '>';
        next if $intag || $c eq '>';
        $i++;
        $out .= "\n" unless $i % $len;
    }
    $out .= "\n";

    is ($out, $x->{want});
}
[download]

This isn't robust (and is rather C-ish for my taste) but it serves to illustrate this approach in general terms. Have fun with it.

Update: edited source for improved generality.