For now, we only have 1 example of a proven sector for the production 1GiB sector size, which means certain extrinsics' weights (declare_faults, declare_faults_recovered, terminate_sectors) aren't accurately scaling with the number of sectors.
When we have test data for more sectors (or figure out a good way to test it without them), the benchmarks need to be updated with n: Linear<1, N> parameters and re-run.
As the code is written right now we'd need to run separate PoRep proofs of the 1GiB sector data, which each takes 20 minutes, and right now DECLARATIONS_MAX == 3000 and MAX_TERMINATIONS_PER_CALL == 32 which means about 3 solid years of proving.
Once we figure out what would be reasonable values for those constants and how much would even fit in a block (using non-scaled weights as an initial estimate), then we can decide on a good limit to run it to and scale it linearly from there.