Although I'm not quite sure why yet, I changed the doctype of this template from XHTML 1.0 to XHTML 1.1 and am sending a Content-type: application/xhtml+xml header to browsers that support it, meaning Mozilla will regard each page as xml and won't show the content if there's an error in the markup.

Which meant spending a couple of hours going through the monthly archives, correcting each entry that triggered an xml parse error. Now all of the 2300+ entries are valid well-formed x(h)tml and I have learned to never ever not end my &'s in amp; or use < instead of &lt; in javascript: hrefs (damn those bookmarklet for-loops).

Remaining Javascript-weirdness resulting from the switch:

• document.write (which I use to generate spam-safe mailto: links) no longer works, I'll have to use a more DOMmy method (explained here);
• The 'hide searches' link above the Referrers disappears after the first click, something to do with my use of innerHTML no doubt.

Update 1: Okay, the 'hide searches' thing was easy to fix; I can simply set .firstChild.nodeValue instead of .innerHTML to make the link text change.

Update 2: My new method of generating mailto: links is as follows: I write out email addresses in this form
<span class="mailto">domain,account,name</span>
which is converted to
<a href="mailto:account@domain">name</a>
by the following function:

function setMailtoLinks()
{
  if (!document.getElementsByTagName) { return; }
  ar_spans = document.getElementsByTagName('span');
  for (i=0; i<ar_spans.length; i++)
  {
    if (ar_spans[i].className == 'mailto')
    {
      ex = ar_spans[i].firstChild.nodeValue.split(',');
      tmp = document.createElement('a');
      tmp.href = 'mailto:' + ex[1] + '@' + ex[0];
      tmp.appendChild(document.createTextNode(ex[2] ? ex[2] : ex[1] + '@' + ex[0]));
      ar_spans[i].replaceChild(tmp, ar_spans[i].firstChild);
    }
  }
}