I only measured what's happening now, of course.
However, you appeared to imply that this was the best that could ever be. It certainly is not the best that ever was because reliable documentation exists for superior past performance.
But you need to assume SOME variability in the times.
Whatever the minimum resulting time is, it's not realistic to assume that every T/O exactly achieves the minimum. My belief is that 5 of the 7 T/Os I observed tried their best to achieve the minimum, and the results varied from 80 to 87 seconds with a mean of 84. And what do you do about the T/Os who don't try very hard because they worry too much about going into BIE? Discipline them because they took 102 or 107 seconds instead of 80 to 87?
There has to be some realization on management's part that variablility is the enemy of greater service levels. There are ways to reduce variability. Operator performance can be improved through feedback. This is what the timers in Paris and Moscow provide.
40 tph operation requires that both equipment and operators perform properly. This probably means that releasing cars with non-functioning motors can no longer be tolerated. It may mean more frequent equipment inspections. It also means that schedules must be planned and executed to the second not the nearest half minute. It also means that falsifying ontime logs can no longer be tolerated. In short, it requires a complete mindset change.
I also think there will be more variability in dwell time than you project. My sample was small, but in 3 out of 10 cases dwell time was seriously increased by events that have nothing to do with crowd levels, and therefore not susceptible to mathematical modeling.
You can put anything into a mathematical model. The real question is how well the mathematical describes the "real" world.
Short of making holding a door, or getting stuck in a closing door, a serious criminal offense, I don't think this sizeable variability is going away.
Train identification is a serious problem. Destination signs are frequently contradictory. This impacts dwell time by forcing passengers: to ask what train it is; wait for the conductor make a barely intelligible announcement, etc. Also variability increases with headway. There would be less variability running shorter headways than longer ones, with the same platform queueing.
The first order of business should be to specifiy service levels including service reliability. The TA's response to poor reliability has been to relax the lateness criteria. Such mindsets must change. That's where I'd start disciplinary hearings - TA management whose mindset does not change.