I don't recall saying 5 GHz was realistic ...
45 nm E8000s are usually good for 3.5-4.0 GHz, in my experience (E8600 can do a little better because of the higher multi and better stepping).
45 nm Q9000s are usually good for 3.0-3.5 GHz, in my experience, though I've had a couple that couldn't get past 3 GHz 24/7 (1 x Q9300, 1 x Q9450).
As for Nehalem vs. Deneb, we'll have to wait and see. I have no doubt that Nehalem will have the overall performance crown, but if Intel locks the non-EE models and AMD can release some competitively-priced quads that overclock well, I would fully expect AMD to make a comeback to some degree in the mainstream sectors.
Nehalem is expected to have a ~40% advantage over current Core 2 Quads in clock-for-clock performance (1.4x) and current Core 2 Quads have a ~10% clock-for-clock advantage over current Phenoms (1.1x). There's no word on whether or not Phenom's performance/clock will increase w/ the new core, though I'd expect at least 5%. Overall, I'd estimate a 50% performance/clock advantage in favour of Nehalem.
Let's say Intel puts a locked 2.66 GHz Nehalem out @ $300 (equivalent to a 4 GHz Phenom) and AMD puts a 3 GHz Phenom BE out @ $200. The Phenom can potentially OC somewhere in the 4.0-4.5 GHz range on air, but requires upgraded CPU cooling ($20-$60). You're looking at possibly slightly better performance while saving ~$50.
Obviously all heresay until the products are actually released, I'm just saying you shouldn't count the Phenoms out just because Intel is kicking AMD's *** in terms of performance per clock.
