QoS and bearer-aware applications - when is best efforts good enough?

I've been in a few discussions recently talking about QoS, IMS, differential classes-of-service over IP networks, and the GSMA IPX.

The usual topic of "guaranteed QoS" vs. "best efforts" always crops up, usually with VoIP or video lauded as the most QoS-able services. There's usually an assertion that customers will be prepared to pay more for assured quality. There's also usually a recognition that Internet or supposed low-quality IP connections are sometimes pretty good.

Obviously for certain applications it is true that customers will pay for QoS - 2 CEOs discussing a merger are going to want a premium connection, and the marginal cost of a QoS-guaranteed IP connection is tiny in comparison with the inherent "value" of their conversation. But for someone wanting to make a 1-hour call to their granny in Australia, it's not so obvious.

Then there is the question of whether any QoS / CoS level is chosen by the application or the network. This fits with a much wider question of whether networks will tell phones how to behave, or vice versa, in the future.

My own view is that that over time, the end-devices (and applications they use) will start to win out. Applications & devices will be able to examine the available wireless (or wired) bearers, and decide which to use based on an algorithm which perhaps compares bandwidth, latency, ownership, QoS, security and other criteria. Another issue here is whether QoS applies to the radio network as well as the core - the best QoS mechanism in the network (or in the IPX) is no use if you're out of coverage.

Surely it makes sense therefore for application to first check whether, at a given time & location, whether best-effort Internet access, or low-tier operator-guaranteed QoS is sufficient. If best effort is good enough, then great. If it is not good enough, the application can then request a better QoS level. There's no point paying for QoS when it's not needed.