It's rare for a large number of mathematicians to collaborate on a single project.  The vast majority of mathematics papers have 1, 2 or 3 authors.

Using the techniques described in http://academia.stackexchange.com/questions/38969/getting-a-dump-of-arxiv-metadata, I downloaded metadata for all the `math` preprints on arXiv, and had a script count the number of authors.  I got the following results:

    251459 papers
    0 authors:      1 papers         0.0004%
    1 authors:      101926 papers    40.5338%
    2 authors:      89630 papers     35.6440%
    3 authors:      43290 papers     17.2155%
    4 authors:      12555 papers     4.9929%
    5 authors:      2949 papers      1.1728%
    6 authors:      700 papers       0.2784%
    7 authors:      216 papers       0.0859%
    8 authors:      102 papers       0.0406%
    9 authors:      40 papers        0.0159%
    10 authors:     14 papers        0.0056%
    11 authors:     9 papers         0.0036%
    12 authors:     7 papers         0.0028%
    13 authors:     4 papers         0.0016%
    14 authors:     2 papers         0.0008%
    15 authors:     3 papers         0.0012%
    16 authors:     2 papers         0.0008%
    20 authors:     1 papers         0.0004%
    22 authors:     3 papers         0.0012%
    23 authors:     1 papers         0.0004%
    28 authors:     1 papers         0.0004%
    37 authors:     1 papers         0.0004%
    60 authors:     1 papers         0.0004%
    67 authors:     1 papers         0.0004%

I took a look at some of the outliers.  Some are ordinary papers but some are other kinds of collective works.  Note that some appear to be cross-listed under other arXiv sections.

- [0 authors: arXiv:1005.0836](http://arxiv.org/abs/1005.0836) (placeholder for a preprint deleted because it was a duplicate)
- [20 authors: arXiv:1011.1764](http://arxiv.org/abs/1011.1764) (regular paper)
- [22 authors: arXiv:1403.2953](http://arxiv.org/abs/1403.2953) (review article)
- [22 authors: arXiv:1501.02155](http://arxiv.org/abs/1501.02155) (regular paper, computer-aided proof project)
- [22 authors: arXiv:cs/9909001](http://arxiv.org/abs/cs/9909001) (report on the results of a workshop)
- [23 authors: arXiv:quant-ph/0701168](http://arxiv.org/abs/quant-ph/0701168) (regular paper)
- [28 authors: arXiv:1009.4134](http://arxiv.org/abs/1009.4134) (regular paper)
- [37 authors: arXiv:1001.4737](http://arxiv.org/abs/1001.4737) (regular paper but seems closer to physics)
- [60 authors: arXiv:math/0406190](http://arxiv.org/abs/math/0406190) (list of open problems collected from a workshop)
- [67 authors: arXiv:1410.0719](http://arxiv.org/abs/1410.0719) (extended abstracts from a conference)

To the asker: There may be some ambiguity here based on what we think of as a "project" and "people involved", and how research is conducted in our fields.  As a mathematician, I think of a "project" as a concerted effort by a researcher, or group of researchers, actively working to resolve a specific mathematical question.  If this effort is successful, it normally results in one or more published papers (sometimes three or four, but usually not dozens), and all the researchers who significantly contributed to the solution will be authors of those papers.  If it's a two-author paper, then there really were only two people working on that project.  (Of course, there could have been other unrelated groups working on the same question, or something closely related, but they would normally publish separately.)  So to me, there is a pretty close correspondence between "project" and "paper".

You might also find this question interesting: http://academia.stackexchange.com/q/34038/1010