The rdkit smarts query has maximum matches set to 1000. I believe that is too low, many simulations have many thousands of molecules and it prevents SMARTs from being used to select a class of them. I'd like to bump the maximum up to 10000. That still prevents wasted computation but is better for large simulations. I have implemented this in PR #3470.
It would also be possible to add a max_matches keyword to the rdkit_kwargs.