Just read it, and although it have some interesting ideas its not really a complete solution. I have my doupts about the architecture they suggest, but even if that works out the paper doesnt describe the most important properties of the technique, the datastructure and algoritme to support this approch. If I read it correctly it also seems to delegate some responsibility for the interest management to the client. Thats like building in exploits =)
On a side note, I study computer science myself, and this paper have many phases where it feels like the "I dont understand this, lets toss out some fancy word and skip over it" solution
Edit: As an example, using aura calculations is technically the most precise answer to telling which objects are within LOS, but you need to do an initial reduction of the set of object you will have to perform this operation on. Also, this method alone doesnt work very well in term of determing when an object should be "removed" from the knowledge of another client. The same moment that updates no longer come in, the observer have to know that it can remove the object, so that it doesnt seem to freeze.