As you say, it makes the stats worthless - to the developer! There's no benefit to a developer in artificially inflating stats, since it just means that they don't have any idea how many people are genuinely using it.
I like the Drupal-style approach of showing downloads vs "reported installs" - perhaps the latter could be automatically calculated based on the number of update check pings from unique IP addresses in the past [insert time period here]? That wouldn't catch every install, but no approach is likely to do that - and it should minimise abuse. The only problem is that you'd obviously only get install stats from 2.3 onwards - but I don't really see any way of getting stats from existing pre-2.3 installs anyway, without requiring site admins to update Moodle.
Depending on whether HQ are capturing IP addresses for plugin downloads, it could even be possible to limit the download stats to unique IP addresses who have downloaded the version in question, which would also help minimise both deliberate gaming and inflation due to accidental repeat downloads.
On the whole, I'm definitely in favour of some kind of stats - both as a developer and as a potential user. It'd be nice to have some kind of indication both of how my plugins are being used (particularly with new versions of Decaf - I ask in the forums for people to test them, but I can't tell whether the lack of responses is because nobody's trying it or because nobody's finding any issues to report!) and how wide the userbase of a plugin I'm considering making use of is.
-Paul