It's worse than that: we want to be able to bring down the FPGAs deliberately if they are drawing too much power. How much data would need to be moved across the bus every time we do this? So far the assumption is that bandwidth is negligable. Edit: 1.6MB can take minutes at standard serial speeds.

Another reason for using the MCU is that it saves space on the FPGA for hashing. The MCU can translate between USB and SPI for example.

Edit2: MCUs can also be reprogrammed without software costing $3000 to produce the bitstream. This makes things like minor protocol changes or calibrating the built-in temperature sensor easy.