> you reach a limit with people getting in and off in the stations, the whole system slows down with trains waiting their turn to enter the station, so real throughput decreases drastically.
Yes, and this would be an even bigger problem for your proposal, that's the point.
> Also, there is a more fundamental problem with reaching high thoughtputs with mass transit: you need to force people out of optimal routes from their distributed destinations and origins, into enforced stations. This makes sense for point to point links like LA to SF, but it's very inefficient when trying to cover a 2D city: the more routes and stations you add, the less you are likely to reach peak capacity of any one link. If you are going for city-wide subway, you will pay much, much more to build large, expensive tunnels that will stay mostly idle at the perifery of the system.
You're hugely exaggerating the problem. You get some inefficiency yes - if we imagine a grid system then many journeys require a change of line and the average journey is 1.27x longer than an ideal path between two points, assuming people don't change their actions at all (e.g. living on the same line where they work). But that's outweighed by the efficiency gains of mass transit - in a city with a decent subway system that longer journey with a change will still be significantly faster than driving.
As for tunnels sitting idle at the periphery of the system, in the real world that's really not the issue you make it out to be, for a number of reasons; housing concentrates along commuting corridors, the places where the population thins out are, by the same token, the places where land is cheap and the train line can run on the surface, and if necessary then trains can turn short or split/join. If you look at the usage levels that subways actually achieve, they're incredibly cheap for what they do.
