The behavior of the '.' marker seems anomalous, very different between gtkAgg and postscript backends, and not very useful:
1) With both backends, only the markeredgewidth property, not the markersize property has an effect on the size. More consistent with Matlab, and I suspect with what other users would expect, would be to have the markersize property control the size, as it does for all the other markers.
2) With Agg, trying to use markeredgewidth to make the size larger produces odd symbols.
3) With postscript, markeredgewidth=20, for example, does make a large circular dot.
What I would prefer for all backends would be for '.' size to be controlled by markersize; the differences between 'o' and '.', as in Matlab, would be that the actual size of the dot would be a fraction of what is given in markersize (so that with a single markersize, there is a clear distinction between a dot and a filled circle), and the dot would have no edge.
I suppose an alternative viewpoint is that a dot really should be a single pixel, so one shouldn't adjust its size; but this goes against the basic idea of having a plotting library in which the result depends as little as possible on the output hardware.
Some backend dependence might be needed: for bitmapped output, one might specify that the dot marker should always include one full pixel, so that dots don't fade away in thumbnails, for example.