Paper available here.
Below are some images using the Oren-Nayer model compared to the Lambertian model. All objects in the image have the same material properties (i.e. all lambertian, or all Oren-Nayer w/a given roughness).
Cornell Science Classroom Model... mdla file:
 Lambertian (roughness = 0) |  Roughness = 0.17 (˜ 10 °) |  Roughness = 0.35 (˜ 20 °) |
 Roughness = 0.52 (˜ 30°) |  Roughness = 0.70 (˜ 40°) |
 Roughness = 0.87 (˜ 50°) |
 Roughness = 1.05 (˜ 60°) |  Roughness = 5 |
 Roughness = 10 |
 Lambertian, w/maxDepth=0 in EON (i.e. interreflections off) |
 Roughness = 0.87 (˜ 50°), w/maxDepth=0 in EON (i.e. interreflections off) |
Simple Sphere... mdla file:
 Lambertian (roughness = 0) |  Roughness = 0.17 (˜ 10°) |  Roughness = 0.35 (˜ 20°) |
 Roughness = 0.52 (˜ 30°) |  Roughness = 0.70 (˜ 40°) |
 Roughness = 0.87 (˜ 50°) |
Comparing the images above, this model seems to create few noticable differences. You can note that the spheres do indeed look flatter. Additionally, in the room up top, you can notice a few diffrences, including on the door frame at the back of the room, and on the three-fold divider blocking part of the door. Also, the room looks noticably darker. I think at least part of this darkening is due to the model.
As far as it goes, I think they accomplish what they set out to accomplish in the paper, though it seems to me for almost all purposes the Lambertian model is good enough. If you have to look hard to determine the differences above, then it seems unnecessary to spend the additional (not tremendous, but significant) time required by this model.
With only one parameter (roughness) to vary, and the fact that the useful range of this roughness parameter seems to be between about 0 and 1, only very limited affects can be achieved, and thus it's probably more worthwhile to use a more flexable and extendable BRDF model (or at least one with more parameters).
Send comments/questions to Chris Wyman (wyman@cs.utah.edu)