Get a clue!
What a bunch of idiots!
First of all, the G4 is a 32 bit processor. No more, no less. The "velocity engine" is able to group 32 bit data into 128 bit "arrays" that can all be processed with a single instruction. This improves the performance in things like manipulating a bunch of pixels - since it can run the same instruction on 4 32-bit (or 8 16-bit, or 16 8-bit, or 32 4-bit, etc) pixels at a time, it can finish a whole lot faster. But don't fool yourselves. You sound like the PC weenies who were wetting themselves over MMX and SSE. For those of you living in a closet (looks like most of you) those are Intel's version of the "Velocity engine." And while the original MMX implementation left a lot to be desired, because it used the floating point registers for SIMD instructions, the Pentium Pro and it's derivatives don't require the context switch and are much more efficient.
AND the P4 is even better at it because it has dedicated MMX/SSE execution units that are capable of taking over floating point execution from the wimpy x87 FPU. So the "velocity engine", while more efficient than Intel's SIMD, isn't anything new. AND I might add that there are a lot more PC applications that take advantage of MMX/SSE than there are Mac apps optimized for the "velocity engine."
What makes the G4 more efficient is the fact that all instructions are a fixed length, 32 bits, it has a huge number of registers available for data and instructions, and it has a massive number of execution units available. Combined with the best branch prediction logic available, it is simply able to do more at one time than any Intel processor.
Think of it this way: You have one hose that can spray 1000 gallons per second. If you want to move more water, you can increase the speed that the water comes out, or you can add more hoses. The P4 tries to push the water through faster, the G4 adds more hoses. The trade off is that the P4 hose will have higher pressure (Mhz) and the G4 has a larger area to push the water through so the pressure (Mhz) isn't as high.
And just to make sure some of you retards understand: I'm MUCH more impressed by the G4 and the performance it is able to wring from even common, unoptimized, code than I am by Intel's latest attempt at "benchmarketing". The 7450 is an awesome design that is only being held back by Motorola's continued schizophrenia.
If you want a formula that will tell you how to "convert MHz" try this one (it's overly simplified so don't bother nit-picking the details, the idiots wouldn't understand anything more accurate): The P4 has 2 1.8 Ghz pipelines that are 20 stages long, the G4 has 4 867 Mhz pipelines that are 7 stages long. Because of the speed differences, the P4 is able to move the same amount of data through it's very long pipelines at about the same speed that the G4 can move through it's much shorter, slower speed, pipelines. The performance difference is due to the more efficient management of code on the G4. There are less performance robbing "pipeline bubbles" on the G4 and their effect is less pronounced because of the short, fat, pipelines.
So, since it's more efficient, the G4 is effectively twice as fast (mostly because it takes 1/3 the time to recover from a stall, branch misprediction, or context switch).