Hi varval,
A good bake of a large scene like that is not just a push of a button. It is a balance between render-time and effort. I doubt that anything useful will come out with just baking it to one object. (Do I understand that correctly) My doubts are also based on your use of Cloner and instances. All in all, if you have a render-farm, that sounds way more attractive.
I had in mind to bake each of the four stores and go from there.
Besides, once the light is baked in, it is super stable (no noise, no flickering), and any adjustment and re-render are fast, except for the eventual re-bake. But also here, once a light is baked, the texture can be just painted. It can often be done even with the Standard ray-tracer engine. Any baked light saves at least one or two iterations of light bounce. With the same GI settings, the results will be much better. If GI is used at all after that.
Yes, I understand, the invested time, but that has sadly no influence on anything.
I hope you find the sweet-spot for your project.
My best wishes for your project
P.S.:
I rendered my backed project with 288 stores in UHD:
Viewport 00:02
Standard 01:56
Physical 06:35
ProRender 18:03 (Diffuse)