A RS-422 method I have used often, is a master that sends out a 3 byte message "who, what, checksum" where "who" is the byte address of a slave and "what" is "your turn". The "what" can also be "please repeat". Rather than confirmation - which takes time - the slaves are supposed to be ready to repeat on request till they receive "your turn". After the checksum, the master switches to receive. The slave must then within 5 mS respond with a 'real' message or a "nothing pending message". If no response started within 5 mS the master will assume the slave dead and poll the next unit. In your scenario (24 bits) - with everything working - it would, at 9600 baud, pan out to: poll 3mS, response 3 or 6mS = 9ms max per poll and response. This would allow 110 messages per second = ~3 per node at 40 nodes. At 19.200 the number of polls would, of course, double, I have run succesfully at 48k and have no doubt that, if required, you can go faster. If you go beyond 31 slaves, you must use half-load 422 interface chips, most of the newer chips are half-load or less. Pay close attention to termination: 3 resistors at one end, one at the other. The most I have driven by this method is 51 80c51's. No, I cannot send you code, all actual code belong to clients.

PS: It is a common misconception that RS-485 is a hardware specification, it is not, it uses the RS-422 hardware specification. The misconception is so widespread that many chips will be listed as RS-485.