Well, my idea of using the FTDI245 chip is the following (and from it cames the 1MB limitation): I would connect the input data (via a buffer - a permanently open 74HC245 would do) into D0-D7; a freely running adjustable oscillator into WR, a flip-flop (7474-like) set by /TXE going high, reset manually pressing a button. That, and of course what's needed to run the chip (crystal, EEPROM, a handful of R's/C's) and that's all.

Basically it is a raw data grabber, nothing less, nothing more.

While the FIFO CAN be filled at a 10MByte/s rate, it would fill up the 384 byte FIFO quickly and then would indicate FIFO-full until the drivers in PC "suck up" the data via USB. My idea is, there is certainly a FIFO-write rate which can be sustained permanently, the drivers being able to empty data as fast as, or faster, than they are written into the FIFO. FTDI promises in the datasheet this rate is 1MByte/s if the D2XX drivers are used (one needs to write a dedicated application for this); or 300kB/s for the virtual-COM-port drivers (a good terminal program would do).
Now this can of course depend on a zillion of factors and for optimum performance some tuning might be needed, but, said roughly, I would free the OS from all other tasks that might interfere, run the PC-side application (a terminal, if the virtual-COM-port drivers are used), and then turn the oscillator's "knob" as long as the FIFO-full LED would stay inactive, or just lights up, and that's it.

Jan Waclawek