In the long term, the way we'll probably get around the RF bandwidth limits (which are physical) will come along some years after we've switched almost entirely to LED lighting. LED lighting systems can carry a piggybacked Internet signal that can be branched building by building, room by room, even more precisely than current WiFi signals, and carrying almost as much data as a fiber. That would clear the spectrum of the download bandwidth of people who are indoors, while user's uploads could remain non-directional RF to the receivers in the lighting system.
LED's are already competittive with street lights (even at current prices and performance, their lower maintenance saves cities money), so the system could be extended to cover the outdoor areas of most cities. Since the streetlights would already be wired up to carry signals, the signals could carry WiFi/cell repeater during the day or to users who aren't picking up the lighting signals, making the current antenna grid vastly denser.
Instead of using WiFi as the model (where you have to connect to a particular wireless network), the system would be a dense, miniturized version of a cell network, where the antennas track users and hand them off to other antennas. In this application, an antenna would try pinging the user optically (via the LED) and if that doesn't work it would try very short-range RF. That would clear almost everyone in cities from the current networks, leaving the existing system for rural areas and highways.
Getting to that level is going to take a lot of time and investment. The communications industry (ISP's, etc) would have to coordinate with LED lighting manufacturers (the signal switching speeds probably requires RGB leds instead of white from blue/phosphor), and the LED controllers would have to include modems and other very sophisticated electronics. Then they'd have to coordinate with municipalities and business to run the signals all through the streelight system, and the entire industry would have to come together on standards and spectrum allocations.
ETA: Oh, and there's also the issue that LED transmitters and receivers aren't able to use frequency selectivity like an RF signal (with LED's, only a few colors are available, so you can't frequency hop up and down the "blue" spectrum). So if you have two rooms where the lighting overlaps, the rooms have to either use different colors (red, green, or blue) to avoid having an area where the signals walk over each other, or time-division multiplexing where "blue" isn't used at exactly the same time by both rooms.
That means the equipment has to figure out how to detect and correct overlaps via switching colors or multiplexing, otherwise the electricians installing the system have to constantly solve the three-color map problem on the fly, which isn't going to work very well.