You didn't check the numbers.
2.26 = Speedup of 7.83.
2.93 = Speedup of 6 and a bit.
That's not possible. They use the same technology.
It is
entirely possible if you map down to what the root cause of the scalability problem is. In fact it makes alot of sense. The 2.26 is a more balanced machine. In other words, there is a closer match to its memory speed than the 2.93.
Typically, in multiprocess boxes you don't see linearly improvements when you can't efficiently parallelize the workload and/or keep the workload fed. The older tech Xeon get worse speed ups because the cores are taking hits trying to get through the memory bottleneck of the shared bus for all 8 cores. The 2.93 has more of a memory bottleneck problem than a 2.26 processor will if they share the exact same memory speeds. More bottleneck problems leads to lower than ideal speed up.
The counter intuitive result would be for the cpu speed to go up 2.93 , 3.0 , 4.0, etc. while keeping the memory system
exactly the same and the speed up number would
not go down at some point (or at least flat). Likewise, keeping the memory system constant and add 2x, 3x. 4x number of cores sharing it and the single-vs-parallel numbers
not go down.
Only balanced systems get linear speedups. Otherwise you run into bottlenecks. Similar to Amdahl's Law (
http://en.wikipedia.org/wiki/Amdal's_law), applying "more" only on one part of the solution doesn't always make it go faster.
The 2.23 system is a better balanced system. The 2.93 system is going to be better at a heavy mixture of both parallel and scalar workloads. The most expensive models are rarely the best balanced ones.
For those who have scalar workload (running primarily just one program at a time that only using a few threads) and are under 8GB (i.e. many games) the 4 cores are even better balanced.