anyway, I've been saying for years now that the whole infinite recursion on identical centroids issue can be solved by simply placeing more than one poly in a leaf node, Kazan has always maintained that the POF structure does not suport this, but I know it does from many years of dealing primaraly with V pofs that do just that.
in the pof's bsp structure you have a block of memory and various offsets in this block point to different nodes, typically you have a swarm of sortnorm nodes, each having five offsets in them, each of these offsets points to a diferent node, in practice, the nodes a sortnorm points to will be either a bbox or another sortnorm. bboxes have no offsets, they are simply followed by one OR MORE polygons which are ended by a EOF marker. sort norms exist in this same fashion, theoretically you could have mutable sortnorms as children of another node, and you could put sortnorms inside bboxes. but the V pofs don't do this, and I'm not 100% sure FS would interpret them correctly (I'm at about 95%). but I do see that the polys of the V pofs in these lists are not cocentric, they typicaly have very similar normals, that's about it, but I think this method would be very useful for forever fixing the dredded infinite recursion error.