AMD seeded out cards as a performance preview, to show that it was serious about regaining respectability in the very high-end space.
Now, being a preview the board is very, very much an engineering sample, such that final performance, we reckon, will be a little better than what we've seen here.
PowerPlay is not enabled on the card, leading to high low- and mid-load power-draw. AMD is aware of the issue and promises to fix it in the final release.
That Anandtech result is spurious, because we bet the HD 4870s in CrossFire aren't being benchmarked properly, that is, with the CrossFire enabled via a user-inputted profile.
If I was a betting man, I'd say that the eventual Radeon HD 4870 X2 will ship with a street price of £299 and performance that's tuned to be just a little higher than two regular 4870s in two-card CrossFire.
It wouldn't be beyond the realms of possiblity for AMD to raise the clockspeeds of the shipping version to, say, 800/3,800, thereby leading to higher numbers.
Running dual GPUs on one board will always lead to instances of sub-par performance when scaling is difficult - Crysis being an obvious example - but it really is the only method for AMD to claw back some respectabilty from the hardcore enthusiasts who have £300 to drop on one card.
Kalniel,
Point taken about the benchmarks. We're working a revised set that will be rolled in as soon as verification of repeatable accuracy is complete.