Clock for Clock AMD PHENOMs seem the fastest option currently. However factoring in HyperThreading seems to put it big in intels favour. I haven't researched why yet, but a hyperthreading core gives nearly 90% of a non HT core on my I7-920.
The only way I can explain that is if you were using two HT cores on separate physical cores. Two HT cores running on the same physical core simultaneously, and both delivering 90% of a physical core doesn't appear to make sense.
Can anyone confirm this behavior?
It depends. If the workload is purely ALU-bound or FPU-bound and there are few instruction interdependencies, then each thread of a HT core should deliver exactly 50% of the performance of a single thread running on it, as they share execution units. IOW HT doesn't help at all increase performance.
However if the workload is memory-bound, if execution units are mostly idle, then each thread of a HT core should deliver about 100% of the performance of a single thread running on it. IOW HT doubles performance.
For real-world workloads, the number falls somewhere in between 50% and 100%. CoinHunter reporting 90% means his mining algorithm is mostly memory-bound.
!? 
