I had the same problem with the Wiki article! But, to be honest I had the same with the last part of your comment. Could you explain your last paragraph in other words because I don't really understand what you mean there?
You want your graph symmetric around various point-in-time axes, so that the scheduling is predictable.
On the distance-time graph, if you've upgraded a line and can run faster trains, you can plot the train with an even steeper slope in the graph. But by doing so, you may miss more of the knots where many different lines in the graph come together. These knots in the graph are moments in space and time when several trains are at the same station around the same time: the time to change trains and make connections.
The system works best when major hubs are roughly 1 hour apart, so you begin optimizing for that fact systemwide, instead of small, few-minute travel time reductions on individual lines. It may be the case that small improvements in travel time on one line don't outweigh the reduction in connection opportunities, so small improvements become harder to justify. You'd have to run certain trains much faster (and spend more money on faster tracks) to maintain the same level of connections in the network.
[1] https://www.zukunft-mobilitaet.net/42868/analyse/integraler-... [2] http://www.itf.hu/index.php/alapfogalmak/integralt-uetemes-m...