Are binary splits always better than multi-way splits in decision trees?

Question

I'm trying to devise a decision tree for classification with multi-way split at an attribute but even though calculating the entropy for a multi-way split gives better information gain than a binary split, the decision tree in code never tries to split in a multi-way. I was also wondering if entropy for binary splits for a categorical attribute can be smaller than that of a multi-way split, because till now all multi-way splits have provided lesser entropy than binary splits (my dataset has categorical attributes only).