Explain how to define response timeouts when many RTU slaves

Related to https://github.com/stephane/libmodbus/issues/83

doc/libmodbus.txt

@@ -58,10 +58,18 @@ The Modbus RTU framing calls a slave, a device/service which handle Modbus
 requests, and a master, a client which send requests. The communication is
 always initiated by the master.
 
-Many Modbus devices can be connected together on the same physical link so you
-need to define which slave is concerned by the message with
-linkmb:modbus_set_slave[3]. If you're running a slave, the slave number
-is used to filter messages.
+Many Modbus devices can be connected together on the same physical link so
+before sending a message, you must set the slave (receiver) with
+linkmb:modbus_set_slave[3]. If you're running a slave, its slave number will be
+used to filter received messages.
+
+The libmodbus implementation of RTU isn't time based as stated in original
+Modbus specification, instead all bytes are sent as fast as possible and a
+response or an indication is considered complete when all expected characters
+have been received. This implementation offers very fast communication but you
+must take care to set a response timeout of slaves less than response timeout of
+master (ortherwise other slaves may ignore master requests when one of the slave
+is not responding).
 
 Create a Modbus RTU context::
     linkmb:modbus_new_rtu[3]