I'm not really in favor of this for two reasons:
1. Blanking the last three digits isn't really enough to protect privacy when there's a trail of information here. Perhaps you see me editing some respectable topic several times, and then you see me visiting a bunch of pages. You'll be able to know what I viewed.
Since there's nothing scandalous about anything on Wikipedia, we don't need to be paranoid to great extremes about this sort of thing, but we don't need to publish web server logs to the general public, either.
(perhaps this could be solved by removing the ip numbers completely, but this precludes some kinds of analysis, as does last-three-blanking)
2. These things are big, and getting bigger all the time. I wouldn't really like to have people downloading them on a cron job or anything, since I do pay the bandwidth bills around here. :-)
-------
What I'd prefer is that people who want to write programs to study the logs do so by getting developer access. You could then download sample logs to study on your home machine, but then we could also all work together on custom log analysis scripts to be run on a cron, with reports generated for the general public. (We have to watch the CPU usage, though.)
--Jimbo