I've been discussing this issue on email with Dave, he asked me to post a recap, so here it is.

A little scandal has developed recently in the Italian blogosphere, after a guy figured out how to leverage on a known Feedburner bug in order to climb a popular top 100 list. BlogBabel was using Feedburner, among other services, to update an index of the the most popular blogs, and faking the number of subscriber of a blog feed a blogger was able to get almost to the top (BlogBabel has since stopped using Feedburner's data).

From what I gathered, what is happening is that to count the number of subscribers to a feed, Feedburner uses an information that collective feed readers (such as bloglines or googlereader) put in the referrer: each time it downloads a feed, Bloglines declares how many subscribers they have to it as metadata in the referrer. For example, here's a recent line from my blog log file, where you can see that according to Bloglines there are 382 subscribers to my feed:

65.214.44.29 - - [09/May/2007:05:23:39 +0200] "GET /rss.xml HTTP/1.1" 304 - "-" "Bloglines/3.1 (http://www.bloglines.com; 382 subscribers)"

Sending a false referrer line to Feedburner, claiming that you are Bloglines and there are 2 millions reader subscribed to a feed, Feedburner will add 2 millions to the number of readers, and if anybody is using that data for ranking, that feed will skyrocket to the top.

PS: it looks like somebody has been fiddling with TechCrunch feed.

PPS: inspired by this ongoing story, here's a tool you can use to add fake users to your Feedburner counter.

PPPS: according to Dick Costolo from Feedburner in a comment, the issue has now been resolved.