I'm not sure I understand nirak's point about the forecaster being human. But I will disagree about interpolating between the QNHs. Each forecast QNH is valid for the specified times. There is no interpolating.
Someone probably has more details, but to add something else to the discussion, the temp/QNH section of the TAF format changed a couple of years ago from each temp/QNH being valid for 3 hours starting at the beginning of the TAF's validity, to the present system where the first is valid for 90 minutes and the rest being valid for 3 hours each.
The reasoning, I believe, is that under the previous system, the potential for error was somewhat greater. By way of example, imagine the QNH is decreasing from 1013 at a rate of 1hPa per hour. At the end of each 3 hour block, there is a 3hPa error. Under the new system, the forecast is being made for the central time in each 3 hour block, so the error in that example would be only +/- 1 hPa.
As for the first one only being 90 minutes and the last 90 minutes of 12 hour TAF being unforecast, it's just how it has to be in order to maintain the +/- 1.5 hour system. The final 90 minutes is not an issue for most operations because you need to obtain a forecast within 60 minutes of your departure and few of us are still flying 11 hours later. And if you are, it is likely you could obtain an updated forecast in flight to cover that period.
I disagree with the suggestion about the future being impossible to predict and all that stuff. When it comes to these QNHs, it really is fairly predictable and as a result of that, we are able to fly instrument approaches using only a forecast QNH and the only penalty is adding a 100ft (3hPa) buffer. So with all the risk assessment that goes into instrument procedure design, they have decided they will always be within 3hPa of the correct QNH. Not bad. Even the area forecast QNH is guaranteed within +/- 5.