I just used the Tandy 1000 EX/HX of examples of machines that have additional wait states accessing certain parts of memory. (IE, I've built my own SRAM-based memory expansions for my machines and, like I said, they run about 30% faster running from that memory than the video memory, which is approximately 50% contended by design.)
The point was to ask whether you'd done any testing to determine the practical effect of those additional "6/8" wait states that your method of using the SPI PSRAM for memory has on performance. Considering a normal 8088 memory access is, what, three or four? T-states this actually seems like it would be really expensive.