That's how the Crusoe works, by 'morphing' the x86 code into it's own language (it's based on a Very Long Instruction Word or VLIW). It basically can fit up to 4 x86 instructions in on Long Word. Of course the Intel chip is going to beat the Crusoe clock for clock. Think about it, a chip that has to translate code then run it or a chip that can run code directly? No contest. The beauty of Crusoe is that it is able to do the morphing without too much of a performance hit, yet use much less power than the current mobile chips. Another cool thing is that the chip actually gets faster the more times a program is run. It saves the most frequently used instructions so that it doesn't have to retranslate.
Not sure about using the system memory, but if I remember correctly, the Crusoe has a much larger cahce to store instructions.
Most of the benchmarks run on Crusoe chips are the standard suite that all the notebooks go through. It's not really a fair benchmark because one of the selling points is the ability for the processor to speed up by using pretranslated instructions.
I don't really see the Crusoe going after the laptop segment right away. I think it will do better in the internet appliance or PDA segment at first.
I still love the idea behind Crusoe, yet there is much room for improvement. Just remembered another cool feature of the crusoe. Every one here knows about the different steppings of the P3 and Athlon processors right? Imagine the ability to update the processor without having to change hardware. This could be possible with the crusoe because it is a software based chip. The engineers could issue a new version of the morphing software to fix hardware problems. Definately cost effective solution....
Wow that was long. 🙂 Wrote a term paper on several different software translation methods.... Covered these guys for a few pages.