Could you paste the logs for 75ms and 35ms inter transaction delay… I could compare the logs to the to single slave version…
EDIT Got new information on performance: I just got “virtual null modem” to work in my linux pc (see pymodbus for the nullmodem implementation). As modbus slave I used diagslave (./diagslave -m rtu -a 1 -b 38400 -d 8 -s 1 -p none -4 10 /dev/pts/7). Openhab acts as modbus master.
The openhab.cfg is here, items config here.
I have TRACE level logging enabled, and get average performance of 12ms (time between consecutive modbus responses, Response: xx yy ... log from ModbusRTUTransport). Most of the time (11ms out of 12ms) of that goes to the actual reading of the response from the server.
This is of course not really comparable to your situation where there is a real serial line, but shows that there is little overhead with the actual binding side. Compared to real life, the java serial binding openhab uses might not be as performant as it could be, for example.
@mbs38 If you can provide the logs I can find out where time goes with your specific case.