They acknowledged repetition, not high RAM usage. I mean having many different objects would require each object to be stored somewhere and loading them from a HDD on demand is too slow. Keeping them in RAM the whole time requires too much RAM. I really don't know how they have solved that problem.
I watched the 24 minutes interview that winkio posted. They are backing up their claims, but they are not really answering any of the problematic questions. He mentioned sorting at one point so they are probably using some sorting algorithm for their search algorithm. I just can't think of a searching algorithm that could this easily find the proper pixel and put it in a bitmap and still allow 20 fps. Even octree isn't that fast and it's complexity is log2(n). He mentioned 21 trillion atoms which would be impossible without repetition (even for an atom that needs just 1 byte, this would require 21 TB of RAM). So this can't be all solid memory. He has objects in some kind of format that are then positioned somewhere. But that clashes with an organization where a search algorithm would be useful. In any case, their search algorithm is the very backbone of this new technology. Without it, nothing would work. Because I can't think of a remotely realistic model for a search algorithm for this, I remain skeptical. I would guess that it's based on a special mathematical model abstract that isn't used in traditional physics and math. e.g. 3D projection is already handled with transformation matrices. These guys probably have a whole different approach. He even said that he had to reinvent rotation and scaling which pretty much lets me come to the conclusion that he isn't even using matrices. That's the only way I could think that something like this would be possible.
Don't get me wrong, I am impressed by what they have done. But they leave up way too many open questions for me to be convinced. I'm not saying it's some kind of trick or whatever, but some things simply don't make sense and until I see the finished product, I will remain skeptical. If they are really doing what they claim, then they simply have to have done it in a way that hasn't been done before (as I mentioned matrices in 3D graphics nowadays). I understand that they don't want to release any data into the public or otherwise some big companies could simply replicate what they are doing and even be done before they are (they are just 9 people after all, imagine what a 90 people team could do). Heck, I even want them to succeed in whatever they are doing. But until they come clean, I will stay skeptical.
@Sub: No, in the 24 minutes interview, their representative states that he isn't using raytracing. Raytracing is way too slow to be able to squeeze this kind of performance out without even using a GPU. In fact, it's none of the known techniques. This is why I believe that they aren't even using traditional methods of 3D rendering. They can't even use traditional methods because their system is completely different.
EDIT: Now I remembered something else that's weird. In the interview the reporter asks about bump mapping. With this method there is no bump mapping, you already have atoms, but the guy said that they will implement bump mapping which doesn't make any sense. Either he didn't hear the question right, he didn't pay attention to the question or he didn't want to answer it because he'd either reveal too much information or something harmful to their claims. In any case, these guys are hiding something, be it a new 3D rendering technology from the ground up or the truth behind a skillfully manufactured scam. I hope it's the former, because to me everything indicates this.
EDIT:
@winkio: You said you know somebody who's working on something similar, right? Am I right with my assumption that this is a completely new approach to rendering 3D space? That would be some massive out-of-the-box thinking, don't you think?