So, first thing that's helpful to know in answering is that there is actually a schedule for the subway. For example, here's the one for the A line:

http://web.mta.info/nyct/service/pdf/tacur.pdf

At least back in 2012 or so, when I cared, the A train did keep reasonably close to this schedule (effectively genuinely on it when it wasn't rush hour). It definitely does not keep to this schedule now.

Now, do New Yorkers actually know this schedule? Well, sometimes, actually: while trains do come every few minutes in Manhattan, any specific train comes much more rarely, up to about every 20 minutes for the A train I linked above. This matters if you can't afford to live in Manhattan, or even if you live far away: for about a year, I commuted regularly from 181st St. to Fulton St. on the A, and you better damn well believe I had a specific train I wanted for that trip.

But even for those who don't, if the train gets too far off schedule, you start getting delays, because the trains cannot handle the throughput. Maybe for those commuting within the Midtown area, you know that a train will come every 3-5 minutes, but if the train that would normally take you is still 15 minutes away, then you've got at least a 12 minute backup of extra people who are going to be on that platform. Not only does that mean you won't manage to get your normal train (because you won't be able to reach the platform); it also compounds, since it'll cause people to enter and exit the trains more slowly. There are also many times when trains join or split from sharing the same track, and scheduling delays can change a smooth swapping of a switch back and forth to a traffic jam as controllers try to figure out whether it matters more to give the E or F priority on such-and-such a segment given that the nearest D and C trains are in such-and-such a place. You get this same kind of thing on your compute with disk seeks: it may not seem like a big deal if a read request shows up a millisecond late due to bus contention, but if it means that the drive head has just moved past the data you want, you're gonna have a long delay while it sweeps back, and that in turn may mean that other requests get delayed and so on.

So: yes, the schedule matters even if you don't know the schedule; but it still causes problems even if you want to think of it purely as "bandwidth."