I'll try to draw this back, although most opinions seem solidly anchored, to discussion on Conroe. I'll start by saying I heartily agree that this chip is still not in anyone's hands and until it is this discussion should remain at the amusement among ourselves level really.
2 chips on the same die is "more" efficient. But as you yourselves who own AMD chips know, the real test is what does it do for the average person in real life on real apps? There isn't a lot of comparison to work with, but honestly I've never seen the single packaging as anything more than a bullet point on a marketing list. The speed advantage it brings at the speeds at which these chips operate for a desktop machine is really meaningless. I say that not to be mean, but to help us drop from discussion things that are really just bullet points and of no real effect.
Zram "could" be really impressive. But I have numerous idle curiosities on this regard. Cache has shown several times to not really do tons for AMD chips. The memory controller is so strong, and thus vastly short of saturation 100% of the time, that a higher cache is not guaranteed to bring vast amounts of improvement. The cache on Dothan, as an example of a cache that did worlds of good, helped a rather weak chip (fpu wise) with a rather so-so memory controller, perform decently.
That brings me to curiosity number 2. The efficiency and latency of a cache, if it is to do any good at all, can rarely be determined before it is actually built. Intel learned this to its shame with the cache word size and latency issues in the P4. The size is a much touted number, but the Dothan, again, gained in part because the cache was also extremely efficient and of low latency. Yonah actually steps that down a peg in its design and the performance differences of the two, given the Yonah's other architectural improvements designed to make it better, show that the cache weakening hurt performance a bit. That's why so many folks bought gaming machines with Dothan, even though the Yonah was out in the wild (yes I know there were more reasons than just that).
And thus I come to curiosity number 3. I would assume that we all brought up the K8L and Zram as counterpoints to Conroe. But you have to agree, that we are talking two different timescales of comparable improvements here. K8L and Zram aren't likely to be in silicon before year end, IF even then. I'm highly critical of Intel's timing with Conroe, given all the time they've had to realize the P4's are garbage, but even with that, they aren't going to take till year end to bring Conroe. I'll admit maybe you wanted a "close enough" timing or to compare what was the next big thing for both companies and that's ok if so. But if there is a 6 month timespan between the things you are comparing, and both are future items, perhaps it would be fair to admit that one is likely to come significantly before the other.
Again, with the critique standing on Intel that Conroe has taken way, way too long, and worse, the availability to the public is likely to be scarce even when they do bring it out. (hence why they haven't eol'd the P4 series)
The last point is that it seems near fantasy land thinking to think that the tests (go check xtremeforums if you want more) that have been done on Conroe are not at minimum pointing to a 25% average speed improvement from a similarly clocked K8 of current stepping. That's not entirely peanuts, and it's not just someone tailoring tests. It's pretty hard to muck with superpi for example. Even if you get into wild conspiracy land and say it's only 20% faster clock for clock (and this is quite less than really is the case), you'd be saying a 3.1 clocked K8 will be what stands up to Conroe at 2.67, and it would take a 3.5ghz K8 to withstand a 3ghz Conroe. We don't even have to venture to 3.33 EE editions of Conroe but I'm thinking those are a bit speculative speeds, but the 3.0 has been done from a 2.67 chip. So when we describe the Conroe as faster, at least cede the point that it IS faster, if we are to be comparing our future to future abilities, as well as the fact that this speed really has been demonstrated.