We (the uTouch team) were wanting a solution for low latency through XInput, but we couldn't think of a good way to do it if we had these accept/reject round trips to the server. IIRC, at the X Developer's Conference in 2010 I was giving a talk on how we hoped to architect our gesture stack. I think I mentioned this issue, and Keith Packard mentioned on IRC that maybe we could send events to potential clients before they became the "owners" of a touch sequence.
After many discussions over beers, I went back home and drafted up the touch ownership concept. It has been in Ubuntu's prototype implementation since it debuted in Ubuntu 11.04, and we were able to demonstrate its usage. This gave us all a good feeling that it was the right approach.
Now, we're almost done with our rearchitecture of the uTouch gesture stack, built on top of XInput multitouch. This functionality allows us to have a single process dedicated to gesture recognition, which then dispatches gesture events to clients who want them. The clients can then take them as they are, or analyze them further to see if they match their own filters for gestures. None of that would be possible without huge latencies if we hadn't added touch ownership support :).