Am I correct in assuming that, on a cheap-consumer-grade FPGAs, the pinch point for classic gaming simulation is both quantity and speed of memory access?
Given you usually have enough GPIO pins to drive PSRAM or SRAM that sits outside the FPGA not sure what you are getting to.
It depends on the board design not just the FPGA, the FPGA has block RAM for many purposes, and if you need little of it (like for the system RAM/VRAM) and rely on "carts" for the actual ROMz support then you're good, if you want to use instead non carts to load the ROMz then outside RAM is the solution given size requirements (in the case of PSRAM you can get 55ns http://www.mouser.co...si-pseudo-sram/ and that is ~18MHz access, quite fast for classic gaming direct ROM access, if you like SRAM you can get 10ns [100 Mhz access] but in small sizes http://www.cypress.c...221596/download )
Just to complete the SD2SNES flashcart uses a MT45W8MW16BGX PSRAM chip to hold "teh romz", you can peruse the datasheet here
(and random access time is 70ns)
The turbo everdrive v2 uses a PSRAM chip made by ISSI (v1 used flash instead), the MegaED v1 instead used SDRAM 48LC8M16 (I guess the genesis access patterns are slow enough)