We have deployments with 45 Tbps of traffic and up to 1.5 Gpps (featured in media) of packet rate. In Netflow/IPFIX or sFlow modes FastNetMon scales almost indefinitely. FastNetMon uses our own proprietary high optimised C++ packet parser which ensured top level performance.
In port mirror / SPAN mode FastNetMon can use all your available CPU resources. You will need 2-4 CPU cores for each 10-40G port. Also, you can use kernel side sampling introduced recently and it can lower hardware requirements to 1-2 CPU cores for 40G / 100G interface with decent network card.
As part of FastNetMon Advanced development we do regular load testing in lab environment and last test confirmed that FastNetMon handled 3M flows per second without performance degradation
For speed calculation, FastNetMon uses only single thread by default (limited by single CPU core performance) and it may affect you in case of really big number of IPv4 hosts (tens of millions or more). In this case we recommend using parallel traffic calculation logic as documented here