Speaking as someone who does sometimes do things that require large amounts of meshes, or something that would normally run like a dog with a wooden leg on the first few obvious attempts…
I’d say the main thing is clever thinking.
You’re not going to find an out of the box solution for this, you have to not only explore all possible engine avenues, but also examine your own thinking around the issue.
There will be assumptions you’re making already, that are actually holding you back.
You can ask yourself, what is really necessary, what isn’t really necessary. What can I do that will ‘look’ like that, or have the same effect etc.
Because at the end of the day, end users aren’t going to care about how you did it, and so on. If you really want a wall of millions of meshes, it only has to look like a wall of millions of meshes. In fact that’s the main thing.