Generally the answer will be “it depends”.
If the tiled sprite is static (non moving), the source image size isn’t too big, and it’s only set to a power of 2 total size? It can be more efficient than a bunch of instances of the same sprite.
If the tiled sprite has any changing features at all (location, rotation, scaling, etc), the base image is very large, or GPU memory is at a premium and culling is needed? Tiled sprite will be the worst efficiency out of sprite instances/tilemap/tiled sprites. My understanding is due to the heavy math involved for a tiled sprite ends up being much more tha.
Go with what works for your specific scenario, as there sadly isn’t a one-size-fits-all solution for object type efficiency.