FYI, the tilemap object was never designed to be used with 1x1 tiles. It was designed for each tile to be a reasonably sized independent image.
Using C3, the C3 runtime is faster than the C2 runtime. It appears to sit idle most of the time, but when you click, the C2 runtime spikes the CPU usage up to 30%, but the C3 runtime only spikes the CPU usage up to 7%. So it looks like the C3 runtime is about 4x faster.