It seems Tim Bray made a classic benchmarking mistake. He compared dissimilar things and thinks the result is meaningful. On the other hand, it would be very difficult for most observers to have any idea of how Perl and Java regular expressions are really implemented, so I can hardly fault him for this. I likely would have come to the same conclusion despite my knowing both languages.
<code> <a> <b> <big> <blockquote> <br /> <dd> <dl> <dt> <em> <font> <h1> <h2> <h3> <h4> <h5> <h6> <hr /> <i> <li> <nbsp> <ol> <p> <small> <strike> <strong> <sub> <sup> <table> <td> <th> <tr> <tt> <u> <ul>