The documentation does not state that there is any limit on the amount of
connections per second. This is a bug which needs to be fixed. 

How can you possibly defend any core dump to be correct behavior? It should be
very easy to reproduce this on any system. 

What makes you think that using %c10 will not result in exactly the same
problem in an unlucky scheduling? Your program is simply not robust; it is
broken. 

This is the documentation for the -c% option:
-%cN   maximum  number  of   connections/seconds   (*%c10)  
(--connection-per-second[=N])

Note that nowhere does it mention that N is limited to any particular value
for correctness. If you intended that this program only works for N<=10,
change the documentation. Making any assumptions about the hardware
capabilities of any particular organization is premature also. 

Conclusion, either fix the program or stop misleading people by providing
wrong documentation.