We are talking about a hardware (and not software over hardware, as for example in some games on the SNES console) tile engine with multi-overlapping rectangular (rather than square, when the image of a tall object on the bottom line of the tile overlaps 1-3 lines of tiles higher in the plane) tiles and without a full frame buffer - a maximum of a double linear buffer.
I suppose that it was possible in some of the arcade machines of the heyday of 2D - but my erudition is not enough.
Perhaps, due to the complexity of the implementation, such systems simply did not make sense or rested on some kind of hardware bottleneck? And then the (V)RAM fell in price ...