Regardless of 3d being fake or not, you'd need a system capable of scaling and rotating sprites or layers at least for a Doom lookalike.
Unless you have sufficient hardware to support that solution you have to bend the textures over a specified area (polygon) yourself programmatically. Easy, but the only way to do it fast is on chunky pixels: 1...4 bytes representing a pixel, making it easily addressable from your shader, texture map, etc routine, not to mention direct blind writes to your framebuffer, instead of bit manipulation to get individual pixels. Unfortunately an a1200 sports only bitplane graphics where you have to address each pixel as many times as many bitplanes you use (e.g. 8 times on 256 colours), not to mention each access being bit manipulation, that involves RMW (read-modify-write) operations, unlike chunky mode where you can use just direct W (write) instructions.
The solution was to use:
1, chunky to planar conversion routines: taking an awful lot of processing time on a standard a1200, or using a very low resolution to save on operations.
2, use dedicated hardware, that's why developers demanded to have a c2p hardware in Akiko, the cd32 custom chipset. C= being as cheap as possible managed to screw even that with a "half-automated" solution.
This being the major bottleneck to start with you still have the very low cpu horse power that remains - shared with system activity/dma - to process 3d calculations, AI etc. Not very healthy, something is going to suffer.