Context

Working as a freelance developer, I often made websites completely based on XSLT. In other words, on every request, an XML file is generated, containing everything we need to know about the page content: the name of the user currently logged in, the top menu entries, if this menu is dynamic/configurable, the text to display in a specific area of the page, etc. Then XSL process (caches, etc.) it to HTML/XHTML page to send to the browser.

It has a good point to make it easier to create small-scale websites, especially with PHP. It is a sort of template engine, but which I prefer to other template engines because it's much more powerful than most of template engines, and because I know it better and like it. It is also possible, when need, to give an access to raw XML data on demand for an automated access, without the need to create separate APIs.

Of course, it will fail completely on any medium-scale or large-scale website, since, even with good caching techniques, XSL still degrades overall website performance and requires more CPU serverside.

Question

Modern browsers have the ability to take an XML file and to transform it with an associated XSL file declared in XML like <?xml-stylesheet href="demo.xslt" type="text/xsl"?>. Firefox 3 can do it. Internet Explorer 8 can do it too.

It means that it is possible to migrate XSL processing from the server to the client side for 50% of users (according on browser statistics on several websites where I may want to implement this). It means that those 50% of users will receive only the XML file at each request, thus reducing their and server's bandwidth (XML file being much shorter than its processed HTML analog), and reducing server's CPU usage.

What are the drawbacks of this technique?

I thought about several ones, but it doesn't apply in this situation:

• Difficult implementation and the need to choose, based on the browser request, when to send raw XML and when to transform it to HTML instead. Obviously, the system will not be much more difficult then the actual one. The only change to make is to add XSL file link to every XML, and to add a browser check.
• More IO and bandwidth usage, since the XSLT file will be downloaded by the browsers, instead of being cached by the server. I don't think it will be a problem, since XSLT file will be cached by the browsers (like images, or CSS, or JavaScript files are cached actually).
• Possibly some problems on client side, like maybe problems when saving a page in some browsers.
• Difficulty to debug code: it is impossible to obtain an HTML source the browser is actually using, since the only displayed source is the downloaded XML. On the other hand, I rarely go look at HTML code on client side, and in most cases, it is unusable directly (whitespace being removed).