> That means the people who have the best information about mitigating this issue have a fix that apparently has minimal performance impacts on their cloud workloads

Could Google not just have found a mitigation with non-negligible performance impact, and made a behind the scenes change to the actual hardware resources assigned at each vCPU level to mask the average impact, eating the cost in the short term (while evening it out in the long term by delaying price decreases they would otherwise implement)?