I don't know your objections to your own suggestion,

Basically, it is too directed. That is, it requires parameters to be chosen -- the ratio between clustered picks and non-clustered; the size of clustering subrange; etc. -- which means I would essentially be choosing what to test and thus excluding anything I haven't thought of.

so I don't know if this one is interesting or not

This is very interesting. I particularly like the idea of using images -- whether hand-drawn, or grabbed at random from an image search -- to bias the picking process. It has so many possibilities ...

Eg. grab a random image, process the image with a filter to reduce it to a just points of a particular color or hue; or maybe use a Conway's Life type process to manipulate the pixels until groups of similar hues the reduce to single points; or a dozen other ideas; and then use those points as my dataset.

The only problem with the idea is that it has triggered so many possibilities for investigation, I might never get back to testing the algorithm :) Thanks for kicking down the doors on the box of my thought train!