I think the 3.5 character delay is used principally to provide a timeout in the event that character(s) are lost from a message or you have started listening during a message. All the variable length Modbus RTU messages have a field near the beginning of the message which specifies the number of data bytes within the message, so you can use that to determine when a good message is complete.

Incidentally, a few years ago I worked on several systems which used Modbus RTU and found that none of them adhered to the interbyte timeout specification.