As already stated, the engine use floating point for every transform operation, not double.
You saying the engine needs no modifications for this is bizarre. If the only thing you do is change that node to double precision, when processing that vector the unmodified engine is going to convert the value back to floating point value, making the change to that node meaningless.