I am curious why the time lag was this significant. Mining driving up prices and the lengths miners would go to to get cards would be clear indicators that they were less price-sensitive, meaning Nvidia would be leaving money on the table for years.
The fingerprints for a card running a dedicated mining workload are really clear, and while I've not kept tabs on this for a while, I'm under the impression that GPU-mined coins still typically target the same set of bottlenecks in a manner very dissimilar to what games would do. That's not including the potentially counter-intuitive clocking of core clocks versus memory, if a coin happens to be memory transaction-limited versus math-limited.
Something like a rolling average of activity counters coupled with a duty-cycling of mining-targeted bottlenecks could make this a non-issue for primarily gaming rigs.
While not mining related, I would cite the utilization graph in the following to show how spiky game behavior is, not counting other elements like the various synchronization and graphics calls that would produce stalls or otherwise compromise mining efficiency.
Having a program take some percentage of the GPU's execution time in a 24-48 hour period that leaves it waiting on vsync or API barriers regularly wouldn't be problematic for a gaming rig, rather it's the point of it.
The penalty, like maybe duty cycling the mining workload's bottleneck might not even be noticeable for a game workload as spiky as what is referenced. However, a hashing algorithm that does its best to max out the memory bus 24 hours a day is going to notice, and even a less onerous penalty than halving the hash rate could still be enough to make an economic case for either avoiding a non-mining card or paying more for a mining variant. Other optimizations like odd voltage/clock combinations would stand out as well.