One reason why there may be so many 'broken' servers with 
bad/incorrect responses is here...basically it may have 
been used in some places to work-around IE's hijacking of 
small ( < 512 byte) 404 error pages.

<http://www.caucho.com/support/resin-interest/0005/0373.html>

> I submitted an article with reports to slashdot.org on 
this a while back, it 
> was rejected, no biggie. Basically if you run An apache 
web-server and IE can 
> read in the server headers the word "Apache" it will 
override the error msg. It 
> does this for no other server I tested, Netscape server, 
IIS, yahoo's apache 
> spin-off, and Zope. The only difference is simply one 
said "Apache" in the 
> server-name field and others didnt. 
> If you want to get around it you need to on generating a 
404 or other error 
> page send in the header a 200(OK) signal. So basically 
you need to tell the 
> browser its all okay dokie. Just print out the http-
headers by hand then ie. 
> out.println("Content-type: text/html\n200\n\n Document 
not found, sorry"); 
> should do the trick 
> If anyone works for microsoft, please note you are going 
to a bad place when 
> you die =P 


That said, I've seen another server (IIS 5.0 at 
appnote.stts.edu) that is broken differently.  This server 
returns 403 Forbidden errors until you actually use a 
browser on the site, after which the headers are okay.  
But even with the 403, it still returns the correct 
pages/pdfs to you... ?:-/

17:07:21	4056/4056	-R-M--	error ('Access%
20Forbidden')	text/html	403		html	Pdf
	appnote.stts.edu/thompson/6282.Pdf	I:/web-
archive%20problematic/www.epanorama.net%
2020021228/appnote.stts.edu/thompson/6282.Pdf	(from 
www.epanorama.net/links/serialbus.html)