We observed many performance regressions in BF-43852 due to the upgrade of OpenSSL from version 3.2 to 3.5. One of the regressions we saw is in the agent_connection_timeout workload which measures the ability of the priority port to connect to the server during a connection churn scenario (see this comment). My hypothesis is that this is because the connection handshake is fast in 3.2, causing the  listener thread to be the limiting factor (so the priority port which has its own listener thread is able to connect) whereas in 3.5 the limiting factor is resources for handshakes so having a separate listener thread doesn't help and we observe degraded performance on the priority port. We should confirm this with evidence.