Here is the algorithm for the Mercier calculator (this is the algorithm for the 1999 calculator, but I believe that the algorithm was the same for the 2005 calculator, with the only difference being that data from a different four-year period were used).
http://myweb.lmu.edu/jmureika/track/Mercier/Merc99.html
As you can see, there are a number of unexplained mathematical assumptions in the calculator that I didn't mention earlier, notably these three:
1. "The next step is to assign to each weighted speed (and hence each ranking) an associated score. For the 10 performance (5 men, 5 women), these are
Rank Men Women
5th 965.8 694.2
10th 947.3 673.9
20th 931.8 657.7
50th 910.0 625.3
100th 889.7 597.2"
2. "The 11 (Vwt, Point) pairs are the subjected to a linear fit-- that is, we find the best straight-line equations
Points = A x Vwt + B
Vwt = C x Points + D
which describes the data."
3. "A 'Women's Only' scoring table is obtained by a linear rescaling of the base tables: the associated scores/performances from the men's tables are adjusted by the following equation:
Swomen = ( Smen + 370.23683 ) / 1.10218405
Here, Swomen is the adjusted women's score, and Smen the original men's."
Keep in mind that these same three assumptions are apparently used for every distance from 100 meters to the marathon. It is unclear why these arithmetic adjustments to the data would produce anything close to a useful comparison of men's and women's performances, so -- as you have correctly observed -- if you apply these arithmetic assumptions and end up with bizarre results (for example, that the top dozen marathoners are all women, or that 29 of the top 30 marathoners are women), the most sensible conclusion is that the problem is in the arithmetic assumptions of the model, and not in the training and racing of all of the male marathoners in the world.
A couple more comments:
I'm not sure why "not so far buddy" states that the male population is "normal" and the female population is "non-normal." I assume that what is meant is that the distribution of times for men in various events can be plotted on a normal or Gaussian distribution curve and the women's times cannot. I don't know of any empirical support for that proposition. (Obviously, the mean points for the respective curves will be different, and the standard deviations may be as well.) In any event, I'm not sure that the existence or absence of Gaussian characteristics in the distribution curves has much practical relevance to comparisons between performances of reasonably competitive runners, especially since the data points used in the calculator don't go beyond the top 100 performers in any given year.
I'm also not sure why "not so far buddy" states that the IAAF tables "use a linear comparison." I assume that the contention is that the IAAF tables simply multiply the men's times by a constant to arrive at equivalent times for women. I don't know where "not so far buddy" got that information. (As to whether such a conversion method would be reasonable, I'm not taking any position one way or the other.)