Look at the [Survey of Doctoral Recipients](http://www.nsf.gov/statistics/nsf09317/content.cfm?pub_id=3920&id=2).  This is a large scale survey of US doctoral recipients.  I'm not sure how much they've avoided sample bias.  They have 29,170 employed mathematicians in their sample, of which 

* 17,290 are employed at "universities and 4-year colleges."
* 860 are employed at "Other educational institutions"
* 7,310 are employed at private companies
* 1,050 at non-profits
* 1,290 by the federal government
* 350 by local governments
* 1,020 are self-employed

Of those employed at "universities and 4-year colleges"

* 5,340 graduated less than 10 years before the survey, of whom 880 are tenured,  2,750 are tenure track, 390 not on tenure track, and 1,310 are in positions where tenure is not applicable (not sure what that means...maybe postdocs?)
* 11,950 graduated more than 10 years before the survey, of whom 9,920 are tenured,  520 are tenure track, 730 not on tenure track, and 780 are in positions where tenure is not applicable (again, not sure what that means)

The total number in the sample who got their Ph.D.s more than 10 before the survey was 19,790, so about half of them got tenured positions at universities and four year colleges (of course, that number will go up a lot when you remove people who left academia right after graduating).