I'm not quite sure I follow what you're showing. A 4890 and GTX 285 are not comparable in this sense. The image you provided just shows the cards in Crossfire/SLI with no way to compare them from one to two cards. I don't think you can find how well one system scales without first knowing how a single card scores first.
In the performance score, the 4890 from GPU test 1 to GPU test 2 shows a difference of 4%. The GTX 285 was 3.4% different.
In the Extreme, the 4890 the GPU test 2 was 77.3% of GPU test 1. The GTX 285's GPU test 2 score was 74.4% of GPU test 1. I'm not sure if these scores are comparable in the sense I'm displaying, but this shows the 4890 is scaling better if indeed these numberes are comparable. Also, in the Extreme, the GTX285's score is 135% better than the 4890, but in the GPU tests, it scores 137% and 132%, respectively. This averages out to 134.5%, which is less than the 135%, but still very close.
Edit* These calculations were done with the GTX 285 no Physx.