Rule of thumb for configuring work processes/ RFC handling

Dear experts,

we're running an EWM test system and prepare for go-live. We have18 communication points that send and receive (JCo based) data to/from EWM via RFC. In addition we have around 50 concurrent users (forget this for the moment).

I wonder about the following:

1. How many workprocesses can an application server handle before we should create another instance? The test system is a 64GB, 8 CPU, 8000 SAPS server, the production system a 384GB, 16 CPU, 48000 SAP server. Can I configure it for 120 DIA, 8 UPD, 2 UP2, 5 BTC, 2 SPO? Background: We don't have significant batch jobs running (never see more than two active) and all communication is RFC based with >40 calls coming in per 100ms at peak.
2. We see delays in RFC handling. Trx. SAT shows sometime 10secs or more:
   
   A wireshark trace however indicates that it's the SAP Server who doesn't respond for 10 secs. I wonder if something in the gateway must be configured or checked? We are not clear where the time is consumed. The same call would normally take 300ms (not 10secs).

I would appreciate some ideas from the experts here. We are on Netweaver 7.4 SP11.

Kind regards,

Gunter