So, let us posit that the present pace of order of magnitude increases every four years can be sustained for the foreseeable future. Let us also posit that the IEEE does indeed deliver a 100 Gigabit Ethernet standard on schedule in 2010, giving us a starting point.
As we spin forward, we hit Terabit networks in 2014, Yotabit networks in 2026, and the roof in 2030 with 10 Yotabit Ethernet. That's 10,000,000 Gigabits. Presumably in our lifetimes.
Why do I suggest a roof at 10 Ybps? Current networks are either electric or optical. If we look at the optical side of things, we see that optical networks run in the far Infrared (~1500 nm) or the near Infrared / deep red (~850 nm). Far Infrared runs out of bandwidth at around 182 Tbps, which carries us to 2022. To get more throughput you need to move to shorter wavelengths of light (and we probably will before hitting IR's 'wall' eg our present move to "blue ray" DVDs). Shifting to Blue light gets you up to 564 Tbps. But, that's not enough make the 100 Tbps to 1 Ybps jump at 2026. To get there we need to leap over the visible spectrum entirely into the high end of the Ultra Violet (UV) range. The high UV range (~150 nm wavelength / 2x1015 Hz) gets us past 1 Ybps to 1.4 Ybps -- enough to clear the 2026 hurdle. By 2030 we're up at the very far end of ultra-high UV (~20 nm wavelength, 1.5x1016 Hz). This gets us 10.66 Ybps -- we make the jump to 10 Ybps by a whisker.
What about 2034? There's nothing left in the UV range of the EM spectrum. Beyond that we get into "soft" x-rays. In that land you're no longer dealing with conventional optics, but instead funky metal waveguides. All of the optical work needed to get into UV Ethernet is non-trivial. Transmitting data with x-rays over any distance seems like a genuinely nasty problem. That nasty problem also gets you into ionizing radiation. Ionizing radiation is not your friend. I have my doubts that we'll be welcoming X-Ray Ethernet into homes and businesses.
Hence, I think that there's a roof. I also think that the 'order of magnitude every four years' pace probably can't be kept up. So, IMO, the development of Ethernet speed will probably be spread over more time. But I think that we'll still see Terabit networks by 2020.
Oh, while I'm geeking out completely, lets throw this into the world of Treknology. Transporters. Massive amounts of data. Let's say that you need 32 bits of data to identify an atom and quantify all of those things you need to know about it (charge, spin, etc). This fellow says that there are roughly 7x1027 atoms in a 70 Kg human. At our 'roof' data rate of 10 Ybps it would take 591,511 years to transmit all of the data needed to image a person at the quantum level. No wonder TNG says that their optical computer cores are surrounded by a warp field!
 All of this assumes the very efficient 4B5B encoding presently used. That is, for every 5 state transitions we convey 4 bits of information. 'Light on' to 'light off' counts as a state transition. Up at the very end of things this assumes that we have gained such mastery of light that we can turn a laser on and off in the time it takes to emit exactly one wave of light at a given frequency -- tricky stuff indeed.