Sorry for the doublepost, but I tried the collapse again, and looked more carefully this time.
This is how I've worked out it goes:
a. In a single frame, all natural (cast obsidian) walls fall as far as they are able.
b. They end up at the bottom of the x-liquid pool, full of x-liquid from the pool. The x-liquid they displace is placed above the surface. The x-liquid they held has not yet fallen.
c. At the same time, in that single cave-in frame, any constructed walls/floors deconstruct, and any dwarves around them are crushed.
d. This means that any layers of water or magma are left hanging up in the air where they originally were (unless they manage to fall a single z-level in that frame, but it doesn't look like it).
e. At this point, your dwarven testers are instacrushed [presumably by constructions, since they weren't under any real walls) or, if you're lucky, still up in the air just beneath/in the magma/water. They will never be at the bottom of the pool.
f. In following frames, the magma and/or water falls, creating chunks of obsidian which again collapse.
g. At the same time, the displaced liquid from the testing pool falls back down, again possibly creating chunks of obsidian.
h. Any of these obsidian walls which fall into the pool travel right to the bottom. They would not form a ceiling.
I hate to say it, but the situation looks a little hopeless. I'll check if these effects hold for magma too, but if they do, I can't see any way through.