I've never figured out that whole ninth bit multiprocessor mode thing either. But I have come up with an approach that might work for you.

The fundamental problem, as I see it, is when the slaves send data to the master. If the master transmits data onto the serial line, all of the slaves can receive that data simultaneously. However, when one of the slaves tries to transmit data they are driving a line from, for example, -12V to +12V. But if other slaves are connected to the same line, and their drivers are still holding the line at -12V, well, I believe the technical term for this circumstance is "bad."

The MAX3223 RS232 interface chip has enable pins (FORCEON and ~FORCEOFF) to control whether the drivers are active or in a HiZ state. You can use these controls to turn on the RS232 driver only when a particular slave has something to communicate. If there might be a reason that a slave would need to initiate communication with the master then you will need another approach. But if you can write the firmware such that no slave will initiate communication then this approach should work nicely.

If you just need multiple µ-controllers talking to one another, look into that whole MISO/MOSI thing on the Atmel chips (I don't know if it's on any other derivatives or not).