This work reminds me of the cache rework I did on my HOT-433 v4 board. It came with single-banked 256K soldered on, it was real cache though. The jumper selector header had solder wires, so I also had to solder in jumper headers. Worked fine after the mods, but it makes me wonder why the MB maker didn't simply use DIP sockets and a jumper header. These parts wouldn't even cost $0.25 back then.

Lets say you build and sell 100,000 boards. That .25 suddenly becomes $25000 dollars. Not even including the extra labor of sockets and jumpers. Coupled with the fact that the vast majority of people would never upgrade the cache.

For 100,000 boards, the cost might be more like 1 cent 😀

I suspect it is because the greater the cache amount the tighter the cache timings become (at least in my experience). They wanted to eliminate the risk of people finding out that more cache = slower timings. This seems very motherboard-, FSB-, CPU-, and RAM-dependent. I'm in the middle of a 486 chipset comparison w.r.t this. The results are very puzzling. Double-banked 256K seems the safest bet all around.

I believe too it is about the money, we see this also in our productions for electronics.
The masses in production and small amounts of unneccesary parts add up.
I do not think they could give a .... about timings back then, thats optional and relative to ram used.