Proof of Concept

Orthographic rendered images applied to simple prims, scaled and positioned to match my Blender model. It works about as expected, but I can fix problems in the corners if I can render shadows without rendering the column that casts the shadow.** I may have to start digging into Blender nodes.
Of course the final model will have thin walls and real openings for doors and windows, avoiding evil overlapping z-buffered alpha images as textures. And columns. Preferably sculpties with multiple level-of-detail problems fixed.
See? Spend a year or two in Second Life and even a Baroque antique like myself will start speaking like a native.

**Update. There's a button in the material render pipeline panel for "Only Cast Shadows", Hooray! Blender nodes may lay undisturbed for the moment (better to let sleeping dogs lie).