even microseconds are too large.
More like nano seconds.
FWIW the access time of the best DRAM is still in the 60ns range - and hasn't really changed for 20 years. Other ram technologies have far lower latency (ZRAM is 2-4ns for example) but cost more or aren't yet commercially viable.(*)
_worst_ case latency of PCIe 1.0 is 21ns, PCIe 2.0 is usually less than 10ns
NAND read latency is variable, but even 70us would be inordinately long.
(*) HP has been sampling memristors for a while. At 5-10ns latency I'd take 2Gb DIMMS _NOW_, given how much they'd speed up computational work needed here (the processors spend more time waiting for memory than they do crunching)