The good thing is that you can many iterations even if your graphics card is not great at repeating Geos. Actually self-similar is a little fancy because I'm using three copies of the feedback, and using pix_snap (not snap2tex) because the areas of the texture that only contain the background colour of the window are transparent. So one feedback surface can feedback onto a second.
there is no magic in it. ben was also already postig a patch a while ago, but here again a patch with the basic principle.
its basically the same patch as as the snap2text helppatch, just that the trigger sequence from |t a b| is changed to |t b a| .
best
erich