Gregory Maxwell wrote:
As far as load goes, it's only a question of implementation. A few months back I posted some example performance data for intersections using inverted indexing using the actual enwiki category data: http://lists.wikimedia.org/mailman/htdig/wikitech-l/2006-September/026715.ht...
Even the most evil cases (intersecting two huge categories) ran very quickly. (For example the intersection of GFDL images and living_persons each which has over 100,000 members took 25ms).
I'm with you, just reporting past flamage on the subject. I wouldn't even count on empirical data having much effect... 1/2 :-)
Stan