RadPerf failed requests explanation

[Jim]

I'm using RadPerf to test our servers for 10k auth requests in 10 
requests in parallel. The server used python module and sql, and 
performs around 285 packets/s accepting all the 10k requests but there 
are times when there are several failed requests around 5 to 20. Why are 
there failed requests?

 From RadPerf README file,

Performance testing
-------------------

To see the maximum accepted load that a RADIUS server can handle, use:

   $ radperf -p 100 -s -f test.csv ...

Using larger values for -p will not change the maximum accepted load,
but will change the delay between a request and response.

> "Using larger values for -p will not change the maximum accepted load,
but will change the delay between a request and response"
Why is that?