Its true that a specific maximum number can not be given, but I would like to know the possible implications of increasing the number of confounding variables. What factors would limit the number of confounds to be included?
1. With too many variables, you run the risk of over-fitting, which is why we have rules of thumb about the number of subjects required per parameter in the model;
2. Some of the potential confounding variables may be what is known as colliders, and you can even introduce bias by including them in the model. Directed acyclic graphs (DAGs) can help sort this out;
3. By using treatment effect techniques such as propensity score matching and inverse probability weighting, we can remove the need to include a large number of confounders in a model.
Need to consider the appropriate adjustment sets required - I recommend starting with a system such as Textor's DAGitty: http://www.dagitty.net/ - and the ability to obtain the necessary data on an appropriate number of subjects for all required adjustment sets.