Perhaps a gif was used on the assumption that gzip wouldn't be able to further compress it.
it is a flawed method. a better method would be to make an xmlhttprequest to an object of known size and compare the time to known times for known connection speeds. this method also has the advantage that you don't interrupt the page being loaded and don't interrupt the user experience since it is run in parallel.
Gzipped data can be incrementally decompressed. Are you sure browsers download the entire page before starting decompression?
(Not that this would make the test much better; I agree it has numerous flaws. Just curious about what browsers actually do.)