Many interesting remarks here. I'll pick up two.
I agree, increasing time limits is a valid concern.
But unexpectedly, I see here another option:
not to increase but to decrease time limit, and nearly equally for everyone.
Less time means generally more power for heuristics, and less to bare performances.
Professional AI contests (e.g. Pac-Man) use something like 40ms per turn (CG could be less radical).
Obviously, compiled languages will always be at advantage, but less so by the simple power of speed.
A better eval or smarter algo would be more critical for the mission.
Given this, a "small conservative time adjustment", as proposed by @acautin, could do the job. For example,
60ms for compiled, 120 for interpreted could be a viable option.
The analogy to sports is actually my favorite one (cf the last CG blog post on e-sports too).
And yes, in many sports you do have categories, e.g. weight classes.
Wording put apart, why not try to do more of that on CG?
We did this for code golfing since the differences between languages were just to blatant.
You could argue that in contests there are already rankings by language.
But you get no CG points and just one t-shirt vs top 50 general.
If the prizes were more "widely" distributed across the categories,
maybe this discussion wouldn't be so recurrent.