If in your simulation individuals could improve themselves as well as their group fit, I'd expect group dynamics to only have any relevance when the number of individuals is very big. Individual improvement leads to faster responses, so you'll need a big amount of diversity to let group improvement make any difference.

That said, I do expect it to be hard. Nature created cooperating species more than once, but not that many times¹. And it had half a billion years of a simulation that was certainly much larger than yours.

1 - Counting placentarial mammals only once, what is very likely correct.