I found an interesting figure in the March 2014 issue of the EMS newsletter, from the article by H. Mihaljevic´ -Brandt and O. Teschke, Journal Profiles and Beyond: What Makes a Mathematics Journal “General”?
See the right column on page 56 in this link. (God help me, I have no idea how to embed that figure in the post. Anyway, maybe it is illegal, so I don’t bother learning.) One can see the “subject bias” of Acta, Annals and Inventiones.
On the left column, there is a graph showing the percentage of papers devoted to different MSC subjects in what the authors call “generalist” math journals (note carefully that these journals are only a small subclass of all journals, chosen by a method that is loosely described in the article). On the right column there is the interesting figure, showing the subject bias. If I understand correctly, the Y-axis is the MSC number and the X-axis represents the corresponding deviation from the average percentage given in the left figure. So, for example, Operator Theory (MSC 47) is the subject of about 5 percent of the papers in a generalist journal, but in the Annals there is a deviation of minus 4 from the average, so if I understand this figure correctly, that means that about 1 percent of papers in the Annals are classified under MSC 47. Another example: Algebraic Geometry (MSC 14), takes up a significant portion of Inventiones papers, much more than it does in an average “generalist” journal.
(I am not making any claims, this could mean a lot of things and it could mean nothing. But it is definitely interesting to note.)
Another interesting point is that the authors say that of the above three super-journals, Acta “is closest to the average distribution, though it is sometimes considered as a journal with a focus on analysis”. That’s interesting in several ways.