Is Internet offload the next bottleneck for mobile broadband?

This article about H3G Austria implementing a "direct tunnel" for data caught my eye today. It fits into an emerging theme I've been seeing about the need to offload mobile data traffic from transiting core network elements unnecessarily. Given that some stats suggest that 95% of all 3G data is going to/from the Internet, it makes sense to "dump" as much of it straight onto a connection to the web rather than routing it through expensive (and capacity-limited) core nodes like SGSNs that add no value.

I heard much the same story about femtocells last week, and I've also heard it mentioned about UMA-type WiFi services, sometimes called "Split tunnel architecture". It makes particular sense for roaming data traffic from PCs, for which there is zero value in backhauling via the home network (and especially for H3G, which doesn't charge data roaming fees to on-net customers).

Now in theory, much of this capability to provide a "flattened" IP network architecture should arrive with LTE, and more specifically its counterpart the Evolved Packet Core (EPC, formerly SAE, System Architecture Evolution). But given the timelines, it makes sense for the more data-centric operators to move ahead sooner.

I have a strong suspicion that offload / direct or split tunnel / (assorted other similar terms) will become the next "big thing" after the current backhaul bottleneck is fixed for mobile broadband operators. There will probably be a few different architectures, which I guess will dovetail with specific operator instances of local (femto/WiFi) radio offload, and macro transport connectivity (owned / 3rd-party).