As Erik suggested - why can you not just use 2 stop bits. As for as the receiver is concerned, there is an extra 1 bit time between characters. In many instances this is an acceptable solution.

As Joe mentioned, the 16550 solution ties up a lot of port pins, that may or may not be a problem for you. apart from that, technically, it is not an issue to interface a 16550 to most 8051s. You have the choice of using the external bus feature of the 8051 (using movx) or bit-bashing the bus interface and saving an extra ic (assuming you have no other external devices like ram or eprom).

Another solution worth considering is to use a bit-bashed (software) uart. No extra hardware needed, just a bit of software and some careful thinking. Receiving using a software uart can sap quite of bit of the cpu's performance so you have to balance your requirements. Expect to consume around 50% of the cpu's performance using a 12 clock cpu and 9600baud. If your application can tolerate such a penalty, then maybe this is the easiest and most cost effective route. There are examples of software uarts here and eleswhere on the web. I think Philips also had some app notes from many years ago describing this method.