However, if we take a given level of trains/buses and operators as a fixed constraint, then we can improve service at the margin by doing things like decreasing variability of arrival times, optimizing connections, opitimizing express/local service, and increasing the amount of information available to riders.

Increasing the number of trains on the tracks is expensive, relative to any of those other things. In particular, providing better information is very cheap, especially if you exploit the willingness of private parties to develop applications that process the raw data for end users (e.g. iphone apps, etc.)

Even more important would be close integration of bicycle technology — and ultmately small transit vehicles — with the MTA and development of applications taking full advantage of its open data policy.