just to test this theory I used tamper data with firefox to remove the Accept-Encoding header and I received ~ 100kb page back and with the Accept-Encoding header present I received ~ 30 kb page but the 30 kb page also had the 2 million null characters. I assume microsoft has a bug in their gzip library or possibly firefox/opera have a bug in theirs.
ok, thanks. so i had tried that. at least on chrome/linux, for me, it's not showing the nulls. at least, i can't see them in view source and saving to a file gives: