The DataFlash connects to the SPI bus, i used a level convertor M74VHC1GT125DF2G for the SO, the Dataflash can tolerate 5V inputs, and a 100K pullup on the 5V for the CS worked for me, i also have the CS connected to the convertor, though if you've only got one SPI item, this won't matter.
I've used the FT232 variants a few times (bm/br/rt), its a decent chip, easy to do in bus power, self power or shared mode. has a 3.3v output too if you choose to use it, i didn't and went with a seperate 3V VREG. It's cheap enough and easy to get a hold of and they have drivers for the major OS'es.
The drivers are the best of the worst, from my POV all USB<>Serial chip drivers are bad in one way or another, they either don't support overlapped io or they're just flakey. The FTDI ones don't support the overlapped, but they've not had BSOD or disconnect problems with them, I have had oddities with them but nothing show stopping. I've mainly just justed them as tx/rx/gnd and then do all the handshaking in software. Typically use them at 115K but have done 230. I'm using a 16Mhz xtal on the AVR, there are tables where the baudrate error margins are listed for different xtal speeds.
Though I'd be wary of the AVR if you're planning to do a run of the boards, I've had nothing but problems sourcing AVR's, and I'm not the only one. I eventually had to cut the firmware i'd written from using the 128K CPU to the 64K CPU because they're the only ones i could find. But i was pretty much at the point where I decided to switch to an ARM or something just because of the chip shortage, granted its affecting a lot of chips, but the AVR's just seem to suffer more.
I suppose you could always do the resistor switch on the CPU too. with a spare output.
I've also considered adding opto isolation, I probably will do this on the next rev of the board, it makes a lot of sense.