rdangerer :
[multiple tempers]
First off, there is a difference in multiple tempers after a cryo and multiple tempers after just a quench to room temperature. If you stop at room temperature then some steels can have unresolved austenite up to 20+%, it is not just a small amount. Thus there is a huge difference in the initial conditions of the two versions of multiple tempers. For cryo, there are those that argue that only one temper is necessary because the maximum amount of martensite has transformed and once this has been tempered, further tempering just adds energy that can't resolve any further martensite transformation and simply induces grain growth and a degredation in steel structure.
Without cryo, I have never seen anyone argue that multiple tempers will degrade the structure. Why do multiple tempers do anything if you don't use cryo? Why can't you just do everything in one go - why not just do one long temper? Well for one thing, it is more productive to do multiple tempers at certain time intervals than tie up the oven for the whole period of time, thus multiple tempers need not be better than one long temper, just be able to get the same result and they would still be preferred. Anyway, it could be a number of factors. The austensite to martensite transformation probably takes place faster at room temperature, this seems likely. And maybe the induced air quenching as you noted speeds up the process. Also maybe the steel "settles" one it goes back to room temperature which induces some stress which is relieved in the next tempering.
In regards to grain refinement, grains are just the small bits of steel, they can differ in composition or orientation from each other. During the heat treat process, the crystal structure of the steel changes forms and thus new "grains" are born. These will continue to grow if the energy is there which is why you can blow a steel if you oversoak it because you will get very large grains formed. Grain size reduction is caused by stopping the growth of the grains that formed. The easiest way to do this is to cause a lot of them to form at the same time.
To understand this, picture yourself in an empty room and walk until you hit something, this length is your "grain size". Now put another person in the room and have both of you do the same thing. As you add more and more people, each person has a much smaller chance of walking a large distance because there are people there for them to smack into, thus the average "grain size" is strongly correlated to the amount of people. For steels this "number of people" is called the amount of nucleation sites, which is just a fancy term for the amount of places the transformation will take place.
This also by the way is why forging a steel can induce a smaller grain size in the steel once it has been heat treated. When you forge a steel you are altering the crystal structure, forcing it into messed up orientations. The fancy term for this is inducing dislocations. These disloaction will be nucleation sites for the transformations in the heat treat and the more of them the finer the grain, as noted in the above. This can also be seen by using the people example. If all the people in the room are facing the same direction, then they can all walk a very long way before they hit something and thus they have a large "grain size". If you went around and randomally turned them about, when they started walking then would quickly hit the people that were turned differently than they were - thus they have a small "grain size".
Back to multiple tempers, austenite has a very large grain size and thus if your multiple tempers are increasing the austenite->martensite transformation, and doing it quickly (high number of transformation sites) you will get the optimum grain size.
-Cliff