blauboad, let's start with the assumption that AMD has good reason to avoid implementing the 32-bit core on 90nm SOI. (Given the problems they had with Barton, this is reasonable.)
Now you have all your production based on the K8 die. So three choices remain:
1. Sell only 64-bit processors and give up any presence in the commodity market;
2. Sell only 64-bit processors and price some of the low enough to compete in the commodity market;
3. Degrade some 64-bit processors and sell them as 32-bit commodities.
I think AMD is opting for the third option. The first option has a problem in that AMD would not provide total top to bottom solutions, so would lose some OEMs. The second option (which I will dub the Sanders option) would kill the ASPs of the whole line. The third option (which reflects Ruiz' reasoning, IMO) keeps the 64-bit franchise as high ASPs and provides product for the commodity end also.
In the second half of 2004, I am anticipating that 64-bit desktop product will be transitioned to socket 939 (and utilize DDR-II) and that all socket 754 product will be 32-bit (and use current DDR).