It seems single threaded performance does not shape up as nicely as multi threaded performance.
We have seen SuperPi, cinebench, deep fritz and CPU mark 99 tests so far.

SuperPi shows impressive gains but isn't superPi quite bandwidth constrained so the reduced latency/increased bandwidth may help?
It looks like ~50% for 32m and ~25% for 1m.

Cinebench 1cpu (32bit) shows ~2% performance gains according to Anandtech, JC's tests look more like ~5%.

Single threaded fritz chess score is ~1,x% faster than penryn.

CPU mark 99 improved by ~9% over penryn.

Performance improvements ranging from 2-9% are not really bad, but neither are they breath taking. However, it may be not that important considering we are in the era of multi threading. I can't even come up with any mainstream single threaded benchmarks (besides games) I could ask you to test, so I could be sure about Nehalem's single threaded performance.

Just wondering whether you guys agree that we may see 0-10% performance gains on some compute intensive, single threaded, but not in any way bandwidth sensitive applications like those above?

Anyway it's not bad at all considering how well Nehalem will perform on average and that most of the transistor budget went into SMT/IMC benefitting only multi threaded and BW heavy applications.

Yes, I know its early alpha silicon, but I am quite sure there won't be much more of a boost than <5% from improved boards and chips.