Turns up the solution for me was this:

use Encode;

require LWP::UserAgent;
my $ua = LWP::UserAgent->new;

my $response = $ua->get("$ARGV[0]");
$_=$response->content;
$_=decode('utf-8',$_);
$result="";
for $r (/<div class="result-text-style-normal">\s*(.*?)<\/div>/isg, /<
+td\s+class\s*=\s*"?multipassage-box[^>]*>(.*?)<\/td>/isg) {
 #($r)=/<div class="result-text-style-normal">\s*(.*?)<\/div>/is;
 $r=~s/<sup>.*?<\/sup>//gi;
 $r=~s/<h[45]>.*?<\/h[45]>/ /gi;
 $r=~s/<(.*?)>//g;
 $r=~s/&nbsp;/ /gi;
 $r=~s/\s*(.*?)\s*$/$1/s;
 $r=~s/\s+/ /g;
 $r=~s/([\x{0080}-\x{ffff}])/'\\u'.sprintf('%04x',ord($1))/ge;
 $result.= ($result?' ':'') . $r;
}
[download]

The main points being: (1) I needed to use Encode and decode('utf-8',$_) my response content, and (2) I replaced high characters with their javascript quoted version. This way I can completely avoid unicode in the database and all encoding transmission issues, and just un-quote them in the AJAX client.