load distribution between sender threads
The current distribution of clients between threads is basically round-robin, which can lead to uneven load on the sender threads, because some "clients" are heavy-traffic forwarders, while others may be just clients that send a single query.
This may lead to uneven distribution of load between threads. This is further troublesome, because even if a single thread's buffer starts to block, it basically stops the entire replay (since timing falls apart).
Additionally, it is impossible to tell the sender "load" of threads, or even of the entire tool.