It's the capacitance on each IC. I'm not telling you whether or not to use buffering, but I am telling you that when you hook more than 4 or 5 IC's AND a cable to your signals, they probaly need buffering, and the only way to tell is to compute the total capacitance and the examine how long it takes the MCU to charge the capacitance in order to tell how long it takes the signal to settle to a valid level. If you use any such signal as a clock, you'd best ensure that the risetime for the clock input is met, too, else you'll need a schmitt-trigger buffer, e.g. 74HC541 or HC244. If a buffer results in a shorter settling time, then you'd best use it. If it computes as "on the edge," the maybe doing without it will work. My attitude is "when in doubt, leave it out," but there are circumstances when it's best to go the other way, such as when you don't know whether there will be more load later on.

Buffers on EVERY signal are definitely indicated whenever more than one board, especially if they're connected with cables, is involved. The clamping diodes on their inputs will help manage any ringing that occurs due to signal path disconinuities, and the like. You might also want to add terminating diodes (74S1053) at the end of the signal path, e.g. at the last expansion connector in a series.

I don't know why you'd want to go with such a small SRAM. I don't know where you'd buy them. The smallest I was able to find about three years ago were 32Kx8's and those were 8 ns parts. They were not very costly. Lately, I've used 64Kx8 parts at 5 volts (61512A-15). Those were acquired as surplus cache RAM's from a former PC vendor's cache stock. They once were found in groups of four on old '486 motherboards.

In over 40 years I've accumulated fewer than a dozen faulty RAM chips of all sorts. Usually it's something else. be sure there's no slack in your wires, BTW. You should never bundle your wires. If you wire "shortest-path" the random interaction will cancel out, only slightly raising the noise floor, while bundling will exacerbate crosstalk, since bundled wires tend to be arranged such that they change at the same time. That causes longer settling times.

RE