I've just made some tests on the FTDI232 chip at high baud rates.

Using a custom routine written in  PowerBasic (CC) , I used that language's own interpretation and interface to the driver to pass arbitrary baud rates to the chip.   With the device set for 8 data,  1 stop bit and no parity, I repeatedly send the character 0x55 in a continuous loop.  This pattern of bits, if sent contiguously (characters are sent LSB first)  should therefore result in a square wave being generated at a frequency exactly equal to half the specified baud rate.

The FTDI232 data sheet states that baud rates are determined based on a 3MHz input to a divisor of (N + M/8, where N = 2 to 16384, plus two special cases on N = 0 and N = 1 for  3Meg and 1Meg rates respectively.   This setting is hidden from my application, and happens transparently based on the baud rate I specify beign passed to the driver via PowerBasic. 

The following results were obtained :

Requested         Frequency

Baud rate           Generated

1M                     500000

2M