If the MCU chosen for the job is the most cost-effective, perhaps it's worthwhile to bit-bang the async comm. 9600 is a pretty common rate, and that's easily bit-banged on a fast one-clocker. If, when you're doing serial communication, particularly if it's rare, you're not doing anything else, i.e. you're in a mode awaiting human interaction, which is often the case, perhaps not in the envirionment where you work, but generally, it's pretty common, you can work around the difficulty, but you're right if it's a major share of the job, or if it's at a really high rate that's difficult to bit-bang with enough precision.

If you have a typical payload of 2 kbytes per hour, you can probably do pretty much anything you want, because, even if it's only 2 kbytes (~ a screen-full) per minute, it's an "occasional" task. If it's >25 kbytes per second, well, that's another story.

If you cannot live with the UART clock for your system timing with its multiple processes, including some that have to be critically timed, then, yes, absolutely, use an external UART if you have to have async comm's and can't bit-bang the thing, or use an MCU that allows use of an external baud clock.

RE