Good gravy. Yes, backwards compatibility is important but this interface sounds like an aberration. (I realize Packard is probably exaggerating a little...)
Would it be possible to have a well-planned step change?
1. make the new interface and the removal of the old an experimental feature that's off by default. Ensure userspace works with both interfaces.
2. Continue like this for two or more releases while it stabilizes. (try to encourage distros to flip the switch early in their dev cycles and see if it sticks?)
3. On the planned day in the planned release cycle, flip the switch everywhere and remove the old interface.
Of course, this only works when there's a single userspace client of the interface, and assumes it's possible to write a userspace utility that can use both interfaces equally well. Luckily both apply here.
I just hate the idea of someone having to waste time rewriting and forward-porting kludgy interfaces that should go away anyway. Or that nasty interfaces would hold things up for years. (OTOH, X is well known for pinning itself into situations like this, it's nothing new!)