What about this:
Side view:
SLS
SXS
S@S
SSS
S = cast obsidian
@ = brave explorers
X = updown stair
L = lava over floor
The bathysphere drops. There are no walls above the dwarfs, so nothing crushes them. There's lava above them, but it *waving hands vigorously* flash-freezes into a new roof before it falls onto the dorfs.
Well in the cave-in test I did, I didn't have lava but I did have a dwarf at the bottom of a column of stairs. The stairs broke free and fell and smashed the dwarves. I believe the X stairs would break loose no matter what was above them and leave an empty square where they were and a dead dwarf in a pile of rubble at the bottom. But I see what you're getting at with the lava - you're thinking the lava might remain liquid during the cave in process, and then flash freeze just after the cave in stops.
It seems 3 things must be true for this to work:
1. A lava tile must not be capable of breaking things underneath it just by falling.
2. The water needs to reach the lava tile and freeze it BEFORE it falls in.
3. BUT the lava needs to only freeze AFTER the cave-in stops.
#1 seems self-evident (falling liquids do not break things like cave-ins do), so the issue is just how long does the lava remain a liquid. That comes down to #2 and #3 and it's a question of timing/order of operations. Maybe an experiment could be devised to test that timing question by itself.
The fundamental problem with a bathysphere is that absolutely anything above an open chamber will break loose and kill things - so if liquid magma did work this way, even if fleacircus's exact configuration won't work, maybe there is some other configuration that can. You don't really need stairs - actually what you want is open space over the dwarf, and as little constructed things as possible between the dwarf and the lava. The trick is how do you suspend a tile of lava without anything underneath it?
My first thought was even if stairs are deadly, maybe a really tough dwarf could survive getting hit on the head with a constructed floor. Because I think what would happen is (provided the timing works right) the floor would fall in just before the magma turns to obsidian.
But then I thought, maybe there actually is a way to suspend a lava tile over a hole without anything at all under it. NFossil's magma dipping corridor post (and the earlier magma mist generator design) made me think of this: what if I used a pump to suck up the tile of magma just before it falls? I'm thinking something like this (side view):
O %%MO
O_mOOOO
OO OOOO
OOdOOOO
OOOOOOO
Top view:
OOOOOOOO
O__m%%MO
O_OOOOmO
O__M%%mO
OOOOOOOO
O = obsidian, d = dwarf, _ = channel, %% = pump, M = magma on top, m = magma below
To keep the magma from destroying the pumps I would need to keep it in a channel. In the diagrams there is a channel with a hole in it over the dwarf. Instead of circulating the magma over the top of the pump, I'd circulate it around the side to be absolutely sure there's nothing over the dwarf except a magma tile.