
Having not coded the BSP technique, I'm not entirely sure that I understand all the details. Having said that...
In essense, the Laidlaw approach appears to be nothing more than a nondatastructure version of a BSP tree, where only the intersecting polygons are split. So, if your BSP code was written such that nonintersecting polygons were detected and the nonsplit polygons were placed into the resulting object, your results should be nearly identical (if not exactly like) the Laidlaw approach.
The Laidlaw approach does create tjunctions because polygons are split along [the arbitrary] planes of other polygons. Once the booleans are completed, I run the objects through an optimization process that removes the tjunctions and collapses polygons where it can. I didn't bother going into this, because that was part of a HUGE piece of very complex code that optimizes generic meshes into the "best set" of convex ngons from a series of ngons and/or tris.
A quick way to remove tjunctions is to simply slide the vertices down the the end of an edge, and collapse the polygon that lies between those vertices. If you do this continuously, you'll end up with no tjunctions. But you'll still end up with vertices in the middle of faces, so ideally, an optimization process beyond this is needed... maybe if somebody sent in an Ask MidNight question about this, I might try to explain the process. <hint hint :>
 MidNight
