Here is another foray into simple quantitative analysis. My record with quick and dirty models like this is like my experience doing algebra on the blackboard in front of a class: miserable. I will proof this one and I certainly hope that Anonymous is still hanging around to put things right.
The number of deleterious dominant mutations depends on conditions (e.g. paternal age) in recent generations, more so the more deleterious the mutation is. If g is the mean age of fathers then, given the model we have been discussing, there are 2g–20 mutations from the father each generation. If the mutations are neutral or recessive or even mildly deleterious they just accumulate for a long time.
Now let us think about deleterious dominant mutations that confer a selective disadvantage s per generation. If u is the fraction of all mutations that have selective disadvantage s, then in the most recent generation the number of this class of mutations is (2g–20)(u), from the last generation it is (2g–20)(1-s)(u), from the one previous to that it is (2g–20)(1-s)^2(u), and so on. The ultimate equilibrium frequency is u(2g–20)/s.
For example consider mutations with selective disadvantage s=0.01. The equilibrium frequency of this is given by this formula, and only 4% derive from the most recent 4 generations.
Now consider more serious mutations, those with selective disadvantage of s=0.10. In this case the last 4 generations contribute 40% of the total number, i.e. about ten times as many. For the even more bad class in which s=0.20 the recent (4 generation) fraction is a whopping 60%.
This is, I believe, the basis of Greg’s remarks that for seriously bad mutations only the mating system, i.e. paternal age, in the last few hundred years matters while for neutral and recessive mutations the numbers depend on a much longer history.