You're right about converting non-ascii characters to numeric references. As for making sure that perl interprets text data from the database as utf8, the Encode module would suffice:

use Encode;

# ... get database text into $text and then:

$text = decode( 'utf8', $text );
[download]

Assuming that the text data drawn from the database really is valid utf8, the "decode" call will set the utf8-flag, and then the regex substitution you showed will work fine.