There are various acceleration structures that can be used beyond the BSP tree to optimize CSG, but they each have their own caveats and bring along their own set of tradeoffs, but with clever structuring of the operations you can work around the slow NxN2 operations.