I have been doing the downloads vs. users analysis with Bitcoin in the past, and found a rough estimate that adoption is 1/3 of the downloads of a mandatory update. (I don't know where the 2/3 go but this is what the data has suggested in the past Smiley).
This part might actually partly assuage one of my earlier points. It would, if there were two mandatory updates in the time frame considered, with download numbers for both.
As for where they go, I'd guess part of it is people trying out, and not sticking with it. I have a few wallets here that I likely won't ever use anymore.
Once mirrors start popping up, it'll become harder to keep track of downloads btw. I don't think there are any yet.
The last mandatory update was on June 10, we only started tracking downloads on July 15. The binaries that everyone is downloading are not nightlies, they are the last stable release, which is the June 10 one. Also, if it was mandatory updates people were downloading, we'd be seeing a lot fewer blockchain bootstrap downloads.
That having been said, we are moving to a distributed download cluster, and will be using a more feature-complete logging platform (since we'll have multiple, geographically distributed instances of nginx), so we'll be able to weed out things that are skewing the stats once we've rolled over to that.