I like the idea. It may be quite a bit more efficient to compute the XP cut-off nightly (when $NORM is computed and votes are handed out) so the DB can throw out nodes as it scans instead of having to sort them all in order to find the top 1000 (is MySQL smart enough to use a limited-size heap for such?). Yes, even the current non-random listings could probably benefit from such a change.
Then there is the risk of people reloading the page over and over to get different samples and the DB load of such. Probably worth trying after I work out the current web server performance issues that caused me to roll back the DB performance stats gathering...
Perhaps Best Nodes should only list best of year, best of month, best of week, and best of day while a separate node lists a random selection of all-time best nodes.