During solving a bug, I notice that in point Extrema code, tolerance check isn't carried over to details, thus boundary points often can't be found in the U V space. I have following patches to correct it.
Please also see http://homepages.cae.wisc.edu/~janeh/OCC/bug.brep
for the surface that shows the problem. pick points along the boundary curve where u is about 0, and trying to find the u, v for the points, you'll see programs gives a wrong (u,v).
Please refer to the following discussion thread for a detail for the problem.