Can't be done. You've probably discovered that Warp Stabilizer does not produce the same results twie let alone from two different cameras.
What happens if I use a replace footage ? Or even better, if I fool AE by switching sources without letting AE know ? What causes a re-analysis ?
I'm actually rendering, and can't test by myself before hours...
You can copy and paste the effect to a new layer without forcing a reanalysis if the source pixel dimensions are the same. This may work fine for steroscopic pairs, in which the cameras are moving identically. I would guess that the subspace warp method wouldn't work well, since it's doing per-pixel distortions, but the Position, Scale, and Rotate method should work.
We didn't intend for the effect to be used this way, but I also can't see why it shouldn't work in this limited scenario. Let us know how it works.
As Todd said, you can copy and paste without forcing a reanalysis or you can duplicate and replace footage but only if you don't use subspace warp.
Here's what one of the software engineers who worked on this feature said when I asked him to take a look at this thread:
"I wouldn't rule out Subspace Warp in this case, mainly because the warp itself is has no high spatial frequencies, so I wouldn't expect exact pixel alignment to be that critical. That said, I wouldn't be surprised if doing any method on both eyes in this way gives you a headache."
Aside from all the technicalities re the Warp Stabilizer, your recorded footage never is 100% congruent in the first place, so even a conventional stabilization would not be transferrable without some intervention. Obviously for that reason companies sell specialised stereo 3D tools/ plug-ins...
Stabilization is now over. I used the perspective option, which I guess is a corner pin, and it worked very well.
The 2 cameras were mounted on the same rail, side by side, at the top front of a big houseboat, cruising on the Seine, thru Paris.
The jitter I wanted to remove was mainly boat pitch (Z rotation) and driving (X pan). As suspected, pitch and pan were really equivalent from one camera to the other, and copying/pasting warp stabilizer once analyse was done, gave such straightforward results that I didn't try other methods.
Thank you everyone for your great help, and sincere congratulations to warp stabilizer dev team!