PixelGuys
Consider greedy meshing or similar.
This is especially interesting for transparent blocks, like water or fog. This could bring down the number of triangles rendered. Greedy meshing might be problematic however due to creating pixel errors at T junctions.
However greedy meshing might need special cases for lighting #61.
Alright, so it seems that overall the fragment shader cost of greedy meshing, at least with the variant I implemented in https://github.com/PixelGuys/Cubyz/compare/greedy_meshing, is quite high. While I was able to offset some of that cost with a depth prepass, overall the speedup is only around 0-20% on my GPU, and it's actually slower on my laptop. So ultimately it's not really worth it, which is why I decided to deprioritize this.
I have tried it, and I have no further ideas for other approaches. Furthermore other solutions for performance and mesh memory usage have been found.