Buenos Aires is about 300 km from Rosario. There's a train that runs twice a day, taking seven hours. The Flutter embedded wireless device, due to ship in 2014 for US$20 with a few kilobits per second and about a 1km radius of communication, could conceivably be thrown out the window of the train at regular intervals to provide a low-bandwidth, low-power communications relay line between the two cities; you would need perhaps 500 of the devices, for a total cost of US$10 000. This sounds like a lot of money but actually I think it's orders of magnitude cheaper than running a conventional communication line over such a distance; and it ought to decrease by another factor of two or three in the near future.
The remaining problems, then, would be to power the devices and to protect them from sabotage.
First, power. Nonrechargeable batteries will only last a year or two, maybe less in the sun. Small nickel-cadmium or nickel-metal-hydride cells might last five or ten years; but the problem remains to harvest sufficient power from the environment to supply the communication line. The device might use 2W in operation, so 2mW of photovoltaic power per device, with adequately efficient power harvesting, might be sufficient.
Second, protection. It's probably not practical to harden 500 such devices to make them impractical to break, but you could probably make them hard to find by concealing them inside of pieces of garbage. This also suggests a way to deploy them from the train without arousing suspicion: every 50 seconds, on average, one member of the planting team throws a piece of garbage (say, a food package) containing a wireless transceiver out the window. The train ticket costs US$2.50, so you could buy 4000 trips (2000 round trips) for the cost of the needed devices; so one person could do this in a practical fashion over some 50 to 100 trips, so two or three months. Ten people could do it in five or ten trips each.
You still need to prevent the devices from being detected by their radio emissions and destroyed, which is difficult if they're basically stationary. The best defense against this, other than spread-spectrum and using legal frequencies (the Flutter devices use a 915MHz frequency that's legal for unlicensed use in the US only) is very low duty cycle and very long periods of time between transmissions — days to weeks. This suggests that all the devices should wake up their receivers periodically at a synchronized time to see if they're being activated, say a few times a day, but not transmit anything unless a circuit is opened.
(Alternatively you could use store-and-forward transmission, but this would result in message latency of months.)
The desire to support solar power, combined with the desire to look like garbage, poses the problem of how to have solar cells that are not exposed to sunlight. This seems impossible, but keep in mind that we only need about 2mW. At the 5% efficiency typical of cheap thin-film cells used for solar calculators and the like, this is 1 cm²; it could be 4 or 5 cm² that's not directly exposed to sun, for example behind a layer of paper.
A perhaps more useful project would connect an off-the-grid location rather than a major city, and could be deployed by bicycle rather than train. There are locations in the Tigre delta, for example, that have electrical power but no broadband internet access.