We can define rules, e.g. 6Gb memory for a target of 240% memtest completion with number of threads equal to number of CPU threads. And rank results on the completion time of set target, differentiating them by DIMM type (e.g. DDR4) and, maybe for competitive purposes, number of memory channels that platform supports.
As a result, we would get a competition system, there certain memory configurations pass this memtest faster, by being actually faster in reading, writing and latency-wise (the idea comes from existing one in RyzenDramCalc)
We already have CPU benchmarks, that depend a lot on mem size and/or rank configuration (e.g. GeekBench, y-cruncher). There is some impact from CPU cores/clocks and cache size/clocks, but it is unknown how much does it matter and it can be tested (clearly, memtest is memory bound by default, so going from 9900K 5.0 Ghz to 7.0 Ghz may make little to no difference).