The studies mostly say that the gender gap appears after primary and before middle school, pretty consistently (as wayfarersgirl pointed out.) I can guess at what's happening: in primary, teachers are starting to come around to the idea that manipulatives are essential, so math instruction often goes through concrete to representational phases, and then into symbolic and abstract. The problem is that the higher one goes, the easier it is to see the symbolic or abstract levels as the starting points for instruction rather than the end points. Even the terminology encourages this; algebra really is just the system of symbols used to describe the math of change, but when you are charged with teaching algebra, how easy is it to start with the symbols and teach kids how to manipulate them? And how many kids are going to miss the connection between those symbols and what they represent? I missed it. I was a good math student, with plenty of problem-solving ability and really good number sense, but the teaching I was offered did not put a priority on understanding how algebra was used and how you could get from a real-life situation to an algebraic equation.

Mathematics instruction has to be grounded in concrete learning and intertwine itself through those stages, back and forth, all the way up. If it's not doing that, the kids whose preferred learning style is to make connections are going to get lost. And yes, the research suggests that that's a more classically female learning style. It can and does lead to the same depth of understanding, even of very high-level concepts - but only if the opportunity to explore the connections is embedded in the instructional methods.