Serving gzipped content directly — bad thing to do

I have my website configured to serve static content using gzip compression, like so:

<link rel='stylesheet' href='http://cdn-domain.com/css/style.css.gzip?ver=0.9' type='text/css' media='all' />

I don't see any website doing anything similar. So, the question is, what's wrong with this? Am I to expect shortcomings?

Precisely, as I understand it, most websites are configured to serve normal static files (.css, .js, etc) and gzipped content (.css.gz, .js.gz, etc) only if the request comes with a Accept-Encoding: gzip header. Why should they be doing this when all browsers support gzip just the same?

PS: I am not seeing any performance issues at all because all the static content is gzipped prior to uploading it to the CDN which then simply serves the gzipped files. Therefore, there's no stress/strain on my server.

Just in case it's helpful, here's the HTTP Response Header information for the gzipped CSS file:

And this for gzipped favicon.ico file:

Supporting Content-Encoding: gzip isn't a requirement of any current HTTP specification, that's why there is a trigger in the form of the request header.

In practice? If your audience is using a web browser and you are only worried about legitimate users then there is very, very slim to no chance that anyone will actually be affected by only having preprocessed gzipped versions available. It's a remnant of a bygone age. Browsers these days should handle being force-fed gzipped content even if they don't request it as long as you also provide them correct headers for the content being given to them. It's important to realise that HTTP request/response is a conversation and that most of the headers in a request are just that; a request. For the most part, the server on the other end is under no obligation to honor any particular headers, and as long as they return a valid response that makes sense the client on the other end should do their best to make sense of what was returned. This includes enabling gzip if the server responds that it has used it.

If your target is machine consumption however, then be a little wary. People still think that it's a smart idea to write their own HTTP/SMTP/etc parsers sometimes even though the topic has been done to death in multiple libraries for pretty much every language out there. All the libraries should support gzip just fine, but hand-rolled parsers usually won't.