That's not a fair comparison, cost/mm^2 does not scale in a linear fashion. The two 340mm^2 chips probably come in at just 10% more than a single 520mm^2 chip because of increased yields for smaller less complex dies.
Cost is a closely guarded secret, and the devil's in the details - it's hard to say anything about it from the outside.
But I wasn't actually talking about cost, I was talking about architectural efficiency. I feel like people are trying to have it both ways: they laud the 5870's perf/mm^2 compared with the 580, but then turn around and laud the 5970's absolute performance. I think to be consistent, if you like the 5970's performance, you should not criticize Nvidia for their perf/mm^2. If you like the 5870's perf/mm^2, you shouldn't be too proud of the 5970's performance crown.