Just ran across Microchip's latest toy, a serial SRAM. I haven't looked at this class much as many didn't have a lot of memory, or were slow. This can do 20MHz SPI, not too bad, but still FAR short of the internal RAM speeds. What could we do, though?
Each memory chip can use a Chip Select, Hold, and standard 3 wire SPI. We can gang the chip select, hold, and serial clock together from each chip. It doesn't look like there's a problem twinning the serial in and out lines together for each chip. what we can do then is get a somewhat inefficient parallel RAM module then, starting at 256Kbytes. Might be useful if you have large blocks of data to transfer and can take some lag at the start and stops. You could talk to this array via a parallel port interface on an MCU, but instead of sending a single 24 bit numeric array for a read, you'd have to multiplex this out to all the IO lines. A little more processing, but probably still much faster than anything else. It's this command addressing that will slow it down the most, but it will work very well for sequential read/writes with, perhaps, less cost than a SRAM of similar price and without worrying about FLASH wearout.