Wether it's a SPARC, Intel, AMD, or Alpha based system, everything that's been said about cache effeciency still applies.
Additionally, my main thrust wasn't about whose multi proc evnironment is more efficient (eventhough these days I have to say that belongs to Alpha/AMD, or is about to), but instead the efficiency of the L2 cache. This is critical! Not just the speed of the cache (wether it runs at cpu speed, or system/bus speed, or something else), or it's location (on CPU or on the board) but how well the cpu deals with a cache miss. In a pipelined CPU, the penalty of a cache miss can negatively affect it's output for 5 to 10 plus cycles depending on the CPU. Now if your CPU is averaging a higher percentage of cache misses, there is going to be a rather negative effect on performance compared to another CPU with either fewer misses, deals better with misses, or both.
It seems more like dudes' bosses have more money than knowledge in this situation and are hearing glamour names or are having smoke blown up their posteriors by greedy salesman.
For the most part, I say a dual CPU system, Intel or AMD would be fine for this application. Of course, SCSI drives would be very important in this application, as well as putting any db's on their own drives.
But then what do I know?