There were Opteron dual socket chipsets which had that arrangement, only one socket with memory controller, and they performed remarkably well considering the arrangement. I think it was more a purist technical leaning on AMD's designers as well as overconfidence of comparative technology strengths that led to no MCM being done although one could have been created in 9 months. Anyway AMD have gone through the pain of their decisions and now is the time to finally reap the rewards of doing a native quad-core which is more efficient coherent inter-core communications which will lead to better scaling from dual-core than Intel's fsb MCM.
Gb - re: "In retrospect it's easy to see that their choice of integrated memory controller precluded doing a dual die approach."
Based on what? Just because they chose not to do it? There's some very weak logic there. There are numerous reasons they could've chosen not to; maybe they didn't turn out to be good ones, but it doesn't necessarily follow that it was a technical incapability.