The original 8051 UART didn't have automatic address filtering. But it did have a mode that can interface with a PC that is abusing the parity to fake the support for a ninth bit. So the original 8051 may have received the 9-bit support to allow it to interface in a multiprocessor bus originally planned for use with manual parity processing.

But with embedded devices, you get incentives to look at current consumption. And interrupts do add response time jitter, so you get incentives to avoid irrelevant interrupts. And before the 100MHz one-clockers, the CPU time consumed by the ISR could not be ignored.

So what started with manual generation/detection of the 9th bit will in the end result in a big incentive to let an embedded processor upgrade from manually processed to automatically processed 9th bit. The PC side could manage just well with manual implementation - at least until we got 16-bit protected mode in the 286, and OS/2 had to abuse the keyboard controller to issue resets as the only way to switch back to real mode making OS/2 machines using 16-bit protected mode just about manage 57600 baud.

My speculation was that the 9-bit mode of the original 8051 was a result of previous use of parity with other UARTs, and one thing lead to another and then the improved UART made the 8051 one of the best processors available for distributed multiprocessor systems.