This is an interesting approach. The article set could be divided into deciles (or some other suitable sized sample). I would guess that the median article size that we currently have is likely fairly constant. The markers for the deciles are likely similarly constant. This is, of course, subject to minor sampling variation. People who use population pyramids let them be indicators of broad changes in populations. Statistically significant variations from key markers would suggest differences in the patterns of participation.
Perhaps one of our more mathematically could do an analysis on "The fractal geometry of Wikipedia participation"
My first guess would be that the article sizes are (from a minimum size that would be the size of an average mini-stub upward) distributed according to a Pareto distribution.
Andre Engels