In both Stockholm and Tokyo, small dwell time delays of at most 5 min make up around 90% of the total delays for commuter trains. To understand these disturbances, we use high resolution data on dwell times and passenger counts from both countries over the last several years. We find that trains in Tokyo are much more congested than in Stockholm, and that the exchange of passengers is modest at most stations in the latter city. In both cities, the range of dwell time delays is quite narrow, with between 40 and 50 s separating the 5th and 95th percentiles. Most delays are thus very small, and even small adjustments to dwell times can make a big difference overall. We find that the data on passengers explain about 40% of this variation in dwell time delays, if we account for non-linear and interaction effects, which is thus a ballpark estimate for how much the exchange of passenger contributes to delays for these trains. We also produce simple, linear models which can be used in practice to assign more appropriate dwell times. To facilitate such improvements, key stakeholders and practitioners have been closely involved with the research in both countries.