Connection distribution to Zabbix Trapper processes is handled poorly

We have had multiple situations where we have agents unable to reliably connect to zabbix server trapper port (Connection timeout) when a few trapper processes are busy processing data. We have seen that these busy processes where distributed the connection and the timeout occurrs while waiting on this processing to complete (meanwhile the other 30 trappers sit idle).

I thought this might be because SO_REUSEPORT was in use but that doesnt seem to be the case. From my understanding the default fork copied socket uses locks to fairly distribute, but it would appear not so?