I've heard even the current (G92) models do, but the feature is only enabled on the Quadro models. If that's true, I'm willing to bet that nVidia will do that again.
Yes, all G80 and beyond will run dual precision instructions, but there is no hardware support, its implemented in macrocode. Question pertains to native hardware support.
The rumors last year were that nvidia would only enable DP FP for Quadro and Tesla products. I hope that's false. The categories of GPGPU applications will expand significantly and if it's enabled in consumer GPUs, I will be very happy.