Based on my understanding, it's not all that smug. I think Intels problems with their memory access architecture (I forget the technical details; IIRC to do with the memory controller being on-motherboard rather than on-chip) have given them a well documented problem in the server side market, and even if they have updated their architecture (not just for the server but eventual dual core and 64-bit markets), they still have a bit of a credibility/image problem to redress with the companies that switched.
Your facts are a bit off. While it is true that Intel's memory controller is part of the northbridge as opposed to part of the CPU, that's not what makes enterprise customers unhappy. The thing that makes them unhappy is one memory controller for all the processors on a single motherboard, which means that the more processors you add, the less memory bandwidth you have for each, causing limited returns.
The competition between northbridge and CPU-based memory controllers has pros and cons on each side:
Northbridge-based:
Pro: Easier to upgrade, as northbridges are replaced far quicker. For VIA-based Socket A, for example, the memory controller had at least five versions (KT133, KT266, KT333, KT400, KT600, and KT880).
Con: Higher latency.
CPU-based:
Pro: Lower latency.
Con: Requires a new socket when the memory controller is changed (See Socket 754, Socket 939, and Socket AM2), and a new revision of the processor.
