Regardless of the disk insertion support, it looks to me like Retro Replay support might be challenging. Have a look here:
It looks to me like there is a lot happening on a write to $de00/$de01. It might be too much when using C for the KFF's ARM cpu to be able to handle in one interrupt.
I think the KFF easyflash and action replay drivers are on the edge of amount of cycles available.
But there is still some optimization possible if you carefully code the C and tune for your gcc version / optimizer. Changing from if-statements to switch-statements can make large differences in worst-case speeds of the interrupt. The C-compiler can create jump tables sometimes, this is much slower than without jump table.
From experience building the KFF soft-kernal driver though, if you spend time optimizing the ARM assembler it will probably be possible to fit Retro Replay with some limitations.
There are some optimizations the C compiler doesn't consider that make quite a difference on the interrupt duration. (like re-using address constants)
The Easyflash disk support not only depends on the 3 $de00 range USB registers of easyflash, but also on 256 bytes of RAM in the $df00 range for trampolining (turning on/off the easyflash rom banks during disk accesses). Combining that again with another cartridge seems very difficult.