I can recommend you consider that the difference
between the fastest speed and
the optimal speed comes down to a number of things. Consider one case,
wherein
we speeded the operation of a multi-computer security system by reducing the
baud rate of the monitor terminals from 9600 to 1200. That reduced the amount
of idle (ENQ/ACK) traffic by 90%.
Moreover, lots of I/O tasks require too much speed if you have to test a flag
before fetching each byte. However, if your processor is running at a rate
synchronized with the transfer, you only have to check the flag once, then
use a
loop tuned for the exact length of the transfer cycle. As Tony points
out, it's
based on the externally imposed rate, and if your oscillator doesn't
synchronize
with that, you have problems.
The nice thing about micros is that there are so many right ways of doing
things.