DX10 cards rendering DX9?

Actually, I don't think the performance jump to DirectX 10 will really mean all that much compared to what the hype says. There will be no magical advantage that a card will have simply by being DirectX 10.

Sure, some have made claims to the unified shader arcitecture... But we've seen from the Xbox 360's R500 that while it does help, it's not a miracle potion... Most games aren't limited by vertex shader power anyway.

And, of course, let's also not forget that it seems vendors are going to pump the prices higher for those top-end-crazy cards. So it's not that the new high-end will be so much greater, perhaps its just that they're adding ANOTHER, even MORE rediculously expensive level above the top one today.

The DX10 cards will be faster but not by a factor of 2, more like 20-40%.

Depending on the G80 rumour accuracy expect the speed of about a GX2 on a single chip, som about 30% increase over GF7900GTX , and about 15% more than an X1950.

The R600 should see a slight bit more performance boost, because it will get some boost from raw fill rate #s from higher clocks and increased pixel specific throughput. But even then expect similar ~30-40% increase over the X1950.
But even then it's not stratospheric, just what we experience with the R9800 -> X800/GF6800 and GF6800/X850 -> GF7800/X1800 moves.

And as for Unified shader performance, drivers shouldn't be a major issue, it should happen automatically within the chip, like it does for the Xenos/R500, and ATi said that the P/V overhead is now gone too, so multi-stage transformation shouldn't leave context overhead that we were all worried about. So to the driver and app, it should simply appear as P/V initially without worry about optimization, what might be done is to increase flow pre determine ratios, where P = 54 and V = 8 to increase predictability, but even then, that would only be when you knowfor sure that that's beneficial.