I know that 100r/s is equal to 1 request per 10 milliseconds.
If you specifiy 100r/s nginx will send 100 requests within the first milliseconds of a certain second.
Once these request are done no request will be handled anymore.
When you would be able to specifiy more granular rate limiting interval you whould be able to get a more continuous load.
see: http://imagebin.org/287286