Not necessarily sure on how much downclocking happens, it is very dependent on how you are running your fabs, what you are targeting, etc. combined with the overall market demand.
Let's say that you need to satisfy demand for a lot of high bin speed parts. You run your fab hot, your yields go down, and most likely you end up with parts that are netting out at their true speed.
If you are driving for volume, you will probably run your fab cold, get better yield but lower clocks.
In the first case, because of the fallout (and the fact that you are looking for higher clocks), you will probably have poor overclockers because each of the die is pushed closer to the limits (less headroom). This is regadless of the marked speed.
However, in the second, you could get more headroom because you might have more than your distribution of mid bins and some of that yield can be pushed lower.
But the to all of this is how many parts did you need to yield at what speed from the wafer?
The net of all of that is that you can never assume that you will get better overclocking from lower speeds. It has just as much to do with the recipe (input parameters) as it does with the distribution curves. It is probably true that you have better odds with lower speed, but you might not see as large of a proportional gain. Your mileage may (will) vary.
Processors are snowflakes, proceed accordingly.