Have a look there. This slide also lists the changes (we already know the process features, listed under the inserted Banias core).
If Dothan would have been scaled down with a ratio of 0.6 (like Opteron), it would be 69 mm² (18 mm² less than it currently has). The listed 2 main changes to the core (besides L2 size) won't make up for the missing 18 mm² ;)
I think, the most plausible reason for this are leakage and power requirements.
Market Data 
Markets 
AI
