Check out this recent Technology Review. In essence, what they are suggesting is that you will minimize processing power in your mobile handset if all the computational work were done in the cloud.
This could be a way to get over the key hurdle of mobile devices: battery consumption. Perhaps the future handheld device is nothing more than a user terminal and a local storage, (for those MP3s). Certainly if the latency issues are overcome, this would be the most logical solution. Prices for handhelds would plummet, battery life would increase and more solution providers could add value to a space which is largely excluded from them; namely the handset itself.
The concept is entirely feasible as we move to 4G networks where bandwidths are large and latencies will be much smaller than they are today.
Certain applications, especially video-related applications become even more realistic as the handset may not need to do anything other than render the stream to the handset. Advertising could get much more localized and targeted as the cloud, being the emulator of the user's device, would be able to insert anything into the stream.
How the billing would work is another issue. Some sort of flat rate would be required because measuring usage down to the cycle count would become onerous. Would people using the cloud in the off-hours get a discount? Would people get two different bills: one for bandwidth and one for cloud usage? These things would have to be worked out.
Now, this might not spell good news for the handset manufacturers, but it would be a huge benefit to the consumer and service providers. It fits the same model Intel was touting 4-5 years ago with respect to a single home computer with multiple terminals throughout the house (or village). It is a model that makes sense but is slow on the up-take, probably because it would mean less computers actually being sold. But the cloud is a different phenomenon. In theory, the cloud will be limitless.