In 2001, Tech TV did a p4 vs. G4 shootout. It found Apple's 867 MHzG4 running about even with Intel's 2 GHz P4 in their attery of tests, and quite a bit faster than the Pentium in some Photoshop tests.
Wxcerpt from http://www.macobserver.com/article/2001/08/28.6.shtml: "But the disparity in chip speed doesn't necessarily translate into better performance for Wintel machines. In fact, the latest Macs are faster than the higher-megahertz PCs when it comes to such tasks as compression and running multimedia software, due to Apple's chip architecture. Unfortunately for Apple, the consumer hasn't always understood this difference, and according to Eric Ross, an analyst with Thomas Weisel partners, the megahertz benchmark has hurt Apple's sales and kept the company from gaining market share.
After this story, Brett Larson, the Mac guru who has long been a part of Macworld Magazine, ran some benchmarks for a 2 GHz P4 and an 867 MHz G4. Those tests showed that some Photoshop tests were about 20-30% faster on the G4. Tests also showed an RGV to CMYK conversion with the G4 stomping the P4. The time savings for this one test more than outpaced the time savings for the other tests, something which Apple capitalized on in the MACWORLD shootout. The tests did not us the dual-800 MHz PowerMacs, nor did they use a dual-processor P4. The testing team concluded that MHz ratings do not matter.
With Altivec, the G4 is capable of some surprising results for certain tasks, but most Mac software still isn't optimized for Altivec, and there is a whole huge class of computing problems that simply aren't vectorizable, making Altivec optimizations impossible. I don't think it's an accident that Moto/IBM/Apple never submitted official SPEC results for the G4 - unofficial results I've seen suggest that it would have been somewhat embarrassing. The G5, on the other hand, is a quite capable chip, and depending on what your needs are, a dual G5 may be a fine replacement for a single P4.