Jeff,

Thanks for the test rig.   Yes, the triangular join isn't pretty.   I chose to take a look at modifying it to have just a single character visitor so that I could at least easily see what the results should be, as the number of different visitors gets so high that repeats are almost non-existent, and in the world of the real web, that's not all that representative of real traffic.   Having done that, the solution runs in 1 minute, 9 seconds.  Not necessarily a problem in a report, but still a lot more resources than I would prefer to spend.   What other technique, however, can deliver the number of removals?   It could be that I just haven't put on my thinking cap quite straight enough, but I'm all ears for technique on this.