Are you asking about projects or papers? (You could measure how big a project is in person-hours, people/project, actual time/project, etc.) What I think of a project may be something spanning many years (possibly most of one's career) with lots of different papers, possibly with different sets of coauthors, or it could be a single standalone paper, or something that never turns into a paper at all. I would guess that most published research papers are pieces of a larger project that the investigator(s) are working on. For example, this is probably true for 2/3rds or 3/4th of my papers, depending how you count. The reason is because math is hard, and one typically can't prove what one hopes right away.
If you just are asking about number of people on papers this is easier to answer. There are two papers of Jerrold Grossman just about math publications you may find interesting, though the information is a bit dated (particularly compared to Nate's answer, but using Mathematical Reviews data, which is more representative than the arXiv):
These study things like trends in the number of coauthors and variation by field. For example, from Table 3 in the first article the average number of authors/paper went from 1.1 in the 1940's to 1.63 in the 1990's, and the number of 3+ author papers went from 1% to 13%. See also Figure 6 of the second article which has a graph. Table 3 of the second article, breaks things down by section classifications (which are subjective, but still instructive). Here is a sample from that table:
Section #authors/paper 2+ authors 3+authors
CS 1.77 53% 17.7%
Combinatorics 1.64 46% 17.7%
Statistics 1.56 45% 8.7%
Geometry 1.34 28% 4.9%
Number Theory 1.32 26% 5.0%
Note these numbers are for papers in between 1980 and 1999, so I suspect the numbers should be considerably higher now, based on the upward trends in Grossman's papers and just from what I personally see in Number Theory. This is also supported by the data from Nate's answer.
(Edit) I just remembered this paper (with 8 authors) which has more recent data than Grossman's (going up to 2009). See Figs 2a and 3a, from which it's clear that there's more collaboration in applied areas. The most recent data in there gives about 2.2 authors/paper for applied math and 1.8 for pure math. (End edit.)
For a comparative study across different disciplines, his paper of Newman is perhaps the most well known (at least to me) study. From Table 1 we see
discipline avg #authors/paper