In theory, the Peltier will create some temperature gradient between the heat sink and the CPU surface, let's say 20 degrees C. If the energy added by the Peltier does not increase the heat sink temperature by 20 degrees, then that should be a net reduction in CPU temperature. I'm sure you realized that already.
I agree, that in practice, it seems like the Peltier doesn't really gain one much. I've tried these Peltier/heat sink combos on our chips at work a few years ago (bought a few from the link above) and found that cooling wasn't any better than just using a similar heat sink with a 60 mm fan blowing on it. And a fan uses a lot less power than the Peltier.
Example for two paragraphs above: For a certain energy (heat) input, and a fixed ambient (air) temperature, a heat sink will reach some equilibrium temperature, at which it radiates/conducts/convects (I'll use "radiate" to mean all three) away heat at the same rate as it is input from the CPU and (if present) Peltier element.
This works, because the hotter the heat sink, the faster it radiates heat. So, increase the heat input to the heat sink and the heat sink's temperature will rise, as the temperature rises, it radiates heat faster, and eventually, it gets hot enough to radiate away heat at the same rate as the heat is coming in.
The downside, of course, is that as the heat sink's temperature rises, so, too, does the temperature of the CPU.
So, if you burn more power in your CPU, the heat sink will still radiate it away, but it will do so at a higher temperature than if the CPU was using less power, and thus, your CPU will experience higher temperatures as its energy use increases.
Enter the Peltier. The Peltier sits between the CPU and heat sink and creates a temperature difference between them. IIRC, the temperature difference is based on the voltage you feed the Peltier. So, if before you had a heat sink/CPU combination which operated at 50C, now, in a perfect world, with the Peltier in between them, you'd have the heat sink operating at 50C, and the CPU would be cooled to 30C.
The catch is that the Peltier also consumes energy, which gets radiated away as heat. So, let's say that we started with an energy input to the heat sink which could be radiated away at 50C. So, CPU and heat sink were at 50C. Now we add a Peltier running off of 12V power. The Peltier uses enough electricity, which it converts to heat, to increase the heat sink temperature to 60C. The Peltier also creates a temperature gradiant of 20C. So the CPU is now experiencing 60C - 20C = 40C which is 10C lower than before.
That's how it should work.
In practice, it seems like the Peltier often increases the heat sink temperature by 20C and creates a gradiant of 20C so that it is a total wash, and a waste of power.
It should be possible to choose the heat sink and air flow so that the temperature increase caused by the Peltier is less than the temperature gradient created by the Peltier. Perhaps Apple did. I haven't had great luck with them though.
Anyway, IBM had some spec. that said that for speeds over 100MHz, the PPC601 needed to be cooled below some temperature and that's why Apple and Power Computing used the Peltiers on PPC601s which ran at 100MHz or faster. I don't know why the 90MHz 7200 has one.