For those who have time to do some coding, the AMS releases tables of how many papers in MathSciNet land in each of the MSC subjects. This should be a more representative sampling of mathematical publications than the arXiv. Unfortunately, the format is a list of every paper, its year of publication, and which classifications it used, so it is not obvious to a human which subjects are the most popular.

For those who don't have the energy to create our own table, David Rusin has a chart where the area of each MSC subject is proportional to the number of publications in that filed from 1980-2000. The classification is too fine to easily answer questions like "Is analysis more popular than algebra" and the time period is not quite what we want. But one can immediately see that any one of Statistics (62), Probability and Stochastic Processes (60), Numerical Analysis (65) and PDEs (35) all dwarf Algebraic Geometry (14), Category Theory (18) and even Number Theory (11).