I'd generally support any idea that has 'conservation of sand volume'. This might include a single side-unsupported block (as might be generated when digging into the side, or by more complex methods) 'spreading out into two ramps (one on original block, one into the one in which it spills), and where two ramps on Z+1 that get 'drawn down' onto the level below might add to the Z=0 ramps in order to create whole blocks again, unless (or until) the blocks below are unsupported to one side and thus 'spill' ramping that way. However, in whatever way it is done, it conserves sand-volume.
Additionally, spilling under raised atom-smashing bridges should ideally prevent the unraising (because you now have structures that could never have had the unraised bridge built across them), if looking for verisimilitude, but this might be taken as written in any general nerfing of atom-smashing that might yet occur.
Interaction of the pseudofluid sand with actual fluids should probably be a matter of evicting the water/magma from the cell (sending it upwards, if necessary) that is fully-filled with sand... I'm minded to suggest that ramped-sand should evict half the fluid, but ramped-cells still have 7/7ths fluid content (when full), so maybe ignore it until it becomes solid-sand.
I generally think that if you were to engineer a sand-fall onto a square that overflows (e.g. sand above a ceiling hatch of a room arbitrarily tall, which is opened) should not bias towards overflowing in any particular direction. We already have in the fluid mechanics a direction randomiser that helps a 6/7ths tile in an otherwise 7/7ths pool be flowed into (and thus 'move' in the opposite direction) from an ever changing direction, each tick, and so this would apply to make the decision as to which way (given a choice) the sand does flow.
I think then that if you were to dump (say) three whole tile-units of sand onto an open space, this would be likely to end up forming a + with a solid centre and ramps on the four peripheries (assuming the "two ramps is a whole unit" conversion). Add another unit and there'd be a ramp on top of the centre and an additional ramp now either extending an arm of the 'plus' or in one of the inner corners, according to how the second 'ramps-worth' of sand dumps off of the 'arm' that it had temporarily upgraded to a solid block. Adding even more sand would tend to produce a pyramid (diagonally-aligned), but would have irregularities.
(Imagine, for the most biased irregularity, that the excess on the top of the 'drop zone' tile continually spilling westward, by pure flukey chance, and that westward tile spilling westward whenever it can, although at various points in this cycle the westward tile would be unavailable to flow into and it'd have to flow north and south (because of the PRNG being particularly biased against east, at this point) until it absolutely had to flow east before it had built up a solid-enough base in order to go back to its inexplicable westward tendency. However, most of the time the randomness would make a vaguely regular pyramid of sand, and all without having to do macro-calculations in order to do a look-ahead assessment of where the next bit of excess sand-pile should spill out onto...)
Soil should be equally capable of slumping, but maybe being capable of a 1.5Z adjacent height differential instead of the nominal 0.5Z height differential of dry sand, and maybe wet sand (or particularly dusty soil) should be given an effective 1.0Z differential, while claggy clay-based soils get even more. The calculations for the higher values of these do rather depend on 'spilling' happening based on what's happening on different Z-levels (as opposed to the original idea that just works out spilling on the same-Z, but then if that's on open space it falls down), but would mean that a sustainable (non-sand) wall could be kept on a particular level by keeping any (multiple of?) tiles'-worth of material in check with walls on the layers above.
And imagine the value of this situation (even with just sand) in protecting against Sapper attacks (i.e. digging enemies). They'd burrow into the side of the fort and through a sand-based 'fileld cavity wall', which would then give a something of a continually-renewing obstacle. However, this also relies upon either convenient surface geology or the ability to treat 'sand units' as haulable and placeable (or at least pond-like deposition).