Re: useful
and an 8% saving in traffic sent externally for the cost of a set of Zopfli boxes on the outgoing lines
The quoted 8% would be applicable to large enough static text files that are actually cacheable.
If the file is cacheable - the client would likely fetch it once only anyway. So there you have it - if you have huge, huge javascript file Zopfli can be of a tiny use (and you can just compress it yourself anyways instead of using proxy). "Tiny" - b/c the clients will use their own cached version instead of downloading it on each request.
Facebook and BCC would have dynamic content mostly, so they need online compression. Zopfli compression algorithm is way way way too slow for online compression. Deflate is fast enough, though.
Side note: sites shall use 'deflate' not 'gzip' - gzip comes w/ crc32 and extra header.
For reference compared to LZMA algorithm Zopfli still ~25% worse in size and still slower processing.
To save bandwidth some browsers shall start supporting LZMA and announce that in Accept-Encoding header - the server would pick the best from gzip/deflate/lzma and server the content w/o wasting tons of cycles and bandwidth on a dated 16bit compression.