mm_flynn:
I agree qualitatively with your points, but would love to read a quantitative analysis. If you're spread out over 2 levels per 1000 feet your chances of collision are 5 times greater than if you're spread randomly out over 10. However, as we're probably not spread out randomly they may be only about 3 times greater.
Papers I've read seem to suggest that we see such a small proportion of the traffic out there, that keeping a good lookout is almost irrelevant to collision risk whilst en-route (flying the pattern, thermaling, ridge soaring it's a different matter). If you see (for the sake of argument) 10% of potentially conflicting traffic when flying at a 'random' altitude but 20% when flying a quadrantal level, this in itself is not going to materially reduce your chances of avoiding a collision.
However, if everybody's flying more or less the same direction albeit at differing speeds, then this will reduce your chances of colliding quite substantially - maybe even by a factor of 3. It seems to me that this question should be 'answerable' by analysis or simulation except you would also need to characterise the spread of airspeeds and feed it in to the model, which would be hard to do.