Performance testing without 'SSL resume' and KeepAlive will probably give a clear picture of platform capacity and should allow a realistic vendor comparison.
That´s why they should match 1:1 the observed TCP connection rate and HTTP request rate.
In real world these numbers will not be the same.
With HTTP KeepAlive an existing TCP connection will be reused to carry multiple requests and related replies. A browser establishes a couple of concurrent connections to a virtual server and tries to resume previous SSL connections (shared secret stored in both peers in cache) to save expensive key negotiations. Whenever a connection handling is internally handed over for offload after the TCP 3-way handshake it will be counted as a transaction, regardless if it is a resume, re- or new negotiation as Aaron already wrote.
Applying i.e. OneConnect allows clientside KeepAlive helps to reduce the TCP connection rate and lowers the number of SSL TPS significantly.