Wait for Ars...
Ars Technica has good '
Processor Review' articles whenever a new architecture comes out. Look for one on the CoreDuo processor soon. (He had FOUR more-detailed-than-former-Intel-employee-me-can-handle articles on the G5 and history of PPC, when the G5 came out.)
But the essence is that the PowerPC became less efficient per clock in recent revisions, while the Pentium-M/CoreDuo architecture became more efficient per clock. And the Core Duo is MUCH more efficient 'per Watt', as Jobs now likes to point out. This derives from the fact that the Core Duo (and the Pentium-M predecessor) was designed from the ground up for good performance with the minimum power usage.
PowerPC's big selling point early on was that it was a 'simple' RISC architecture, where most of the hard work was actually in software. So if you wrote your program efficiently, it would run fast. And because the processor was simple, it wouldn't be hard to ramp up processor speeds. Unfortunately for Apple, Intel had more pressure to increase speed (in AMD) than Motorola or IBM did. The main market of the PowerPC really isn't PCs, it's as 'embedded controllers' in things like Cable modems, DirecTV receivers, traffic light controllers, satellites, and other non-computer devices, where all-out processor power isn't the main selling point. So Intel ramped up processor speeds in their architecture much faster than the PowerPC camp did. (In fact, Intel did this partly by having the core of the processor actually be RISC, with internal translation.) Intel's might in developing new manufacturing processes also helps here. IBM is good, Intel is better. (i.e. if Intel were to take the G5 design, and tweak it for their manufacturing plants, they could probably get 3 GHz out of the design no problem.)
Later, the addition of the excellent 'Altivec' vector instruction set allowed the G4 (and later the G5) to perform amazingly fast calculations on sets of repeating data. (Things like photo, audio, and video processing make heavy use of this.) At the time, Intel's counterpart, MMX, was woefully underpowered. However, over time, Intel introduced SSE, SSE2, and the latest, SSE3; which makes their vector processing at least equal to Altivec, if not better. (I haven't seen a true unbiased comparison between SSE3 and Altivec yet.)
So between clock speed increases going better for Intel, and the leveling of the core architecture; the Core Duo's base architecture SHOULD be about equal to a G5 at the same GHz rating. That means that the dual-core in the Core Duo should be about twice as fast as the single-core G5. Yes, Apple always blows performance improvements out of proportion. They always have. But realistically, a 2.0 GHz iMac Core Duo should perform about equally well (on well optimized native apps,) as a dual-processor 2.0 GHz Power Mac G5.
