Rather than a rotating effect, the sprites jump around at random very erratically.
This must be due to some kind of fixed-point conversion and imprecision, before or during conversion to screen coordinates.
The definition of the OBJ_{X,Y}_Coordinate type may also be a factor.
A more detailed analysis of exactly what changes at the assembly level between versions would be required to fully understand this. Introducing explicit intermediate values with specified fixed-point types would also be useful to find the breaking point.
Considered not a crucial use-case for now, and fixed point demonstrates working just fine elsewhere (e.g. within the Sin/Cos implementations themselves). The demo works fine as-is. When more facilities for interfacing with emulator debuggers are present later on, this will be easier to resolve, too.