The 'peculiarity' of the 7/7ths diggable space and the floor that's not an 8th/7th even when chanelled away suggests that it's lost to the same place as all the other material is lost when carving wall-capable rocks into mugs, bracelets, etc.
Ore and gem tiles I always see as the valuble content clustered/veined/massed into the host-stone matrix (which similarly evaporates on extracting the smeltable/cuttable contents, having not enough integrity of its own to be reconstituted as the original marble/whatever, not even for opportunistically crafting earrings). The floor-slice just extends that, and if a cluster is mid-wall then why should it need to stand out in the thin slice that borders the next block down? And a built-floor is not flush to the natural floor it might replace (or else set flush to the atop-natural similarly built floors it would be indistinguishable with), but raised up just enough to make the standard wall unbuildable by the unimaginative constructors. Whether or not that's enforced by a ceiling too.
I can perfectly understand the data-driven reason why these oddities occur. And conversion (or not) of 'flavoured' flooring after messing with the 'volume' contents clearly arises from a different handling of the struct in memory. Altering any of this behaviour (possibly including an ability to recover proportionate value from the floor-slot) would require a further redesign to the data model, with knock-on effects to the engine. I think it's just easier to accept the current quirks, then relearn any new quirks that crop up after the next revision (Steam-equivalent versions may have the model revised to support additional tile-layering necessities, which may push towards or away from any solution thought 'logical' or consistent by those pondering the current anomolies or otherwise).