I would normally expect the protocol to be designed so that you the receivers state machine have reached the "get EOP character" when the EOP character is ready to send. So if a EOP character is received somewhere else, then the receiver knows something is wrong and must wait for a timeout (if master) or a new 9-bit address byte (if slave). And if no EOP arrives, then the receiver knows it has to wait for a timeout (if master) or a new 9-bit start byte (if slave).

So an EOP is most helpful if having multiple potential transmitters on a cable, and sending messages where some of the listeners do not understand the format of the data so can't compute the expected end of a message. In that case, they have to fight for the right to send first when an EOP have been received or there have been a timeout condition.

Above (unknown format of packet structure) isn't really recommended. So the next alternative for an EOP is as a validity check. Normal 8-bit channels normally manages this quite well with just a CRC-16. After all, it really isn't interesting to know why/where something went wrong. It's enough to know that the last received data has to be seen as garbage requiring a retransmit (unless it is continuously updated real-time data, in which case newer data can be sent instead).

The 9th bit can be a helpful tool besides giving the target address, if desining multi-master solutions, with a need for some form of negotiation. Something a bit similar to token-ring.