well, libagnos-c++ hasn't been tested for performance, so what you describe might be possible, although i really think it's odd.
the unittest i made runs in less than a second, and it performs several calls.
i'll try to get to testing the performance of the various implementations of libagnos, but in the meanwhile,
if you could open an issue about it and attach some sample code with the timings you get, it would help a lot.
by the way, everything should be tested on the localhost. particularly interesting are a c#-to-c# test and a c++-to-c++ test.
thanks,
-tomer
An NCO and a Gentleman