Currently, the gas estimation in mocked mode are underestimated, since the setCode hardhat method is directly setting the implementation code instead of proxy + code at the delegated to address, and so in few edge cases, the behavior of some dapps could differ between testnet and mocked mode.
For better devX, i would prefer to deploy the proxies as well in mocked mode, this way we could get 100% correct gas estimation in mocked mode.
Similarly I think we should use 13 KMS nodes instead of 1 in mocked mode if we want to reproduce realistic gas estimation, even if it makes testing a bit slower