There has never been a RCT which showed covid vaccination reduces mortality. I believe they do but this has not been rigorously shown.
Unfortunately, even though we've had gigantic RCTs on covid none of them showed statistically significant covid-mortality in the unvaccinated population (potentially due to prescreening eliminating people who were at risk of dying), so they weren't able to show any improvement.
Numbers from the population have been suggestive, but unfortunately you can get the same graphs -- an initial spike of reduced mortality with a pronounced fall off in effectiveness -- from a simulation where the vaccine has no effect on mortality at all.
The reason for this is that our stats on the public measure time lagged mortality against current vaccination numbers. People reported dead today actually died a week before and got sick two weeks before that. When the increase in new vaccination is high, you will overestimate the size of the population for the people who died and falsely think the vaccine is more effective than it is. And you end up with your 'per month effect' -- the same statistical error will make the cycle nicely repeat with boosters due to the same population size error for the newly boosted population.
This is the sort of bias that is structurally eliminated in RCT design and hard to avoid in population monitoring.