Hi Telnoi - that is normal, operations like Blend and Sweep go through a fitting process where they insert as many points into the generated surface as needed to make them fit a tolerance of 0.001 units of deviation from the input edges.
This fitting process looks at the shape of the edge curve, but it does not really have a direct connection with the control point structure of the original surface.
There also is not any requirement for an edge to be blended to be aligned with the natural UV layout of a surface, like for example here I'm blending between 2 trimmed edges and you can see that the control point grids of the surfaces involved don't really have any directional relation to one another at all:
This is kind of related to how NURBS surfaces work with "underlying surfaces" and trim curves, where you can have a trim edge that is on a surface but does not necessarily align with the control point structure of the surface. Some more information and illustrations to explian that
on this FAQ answer here.
Basically a generated surface is good and properly formed if it is close enough to the original inputs so that it can be joined into a having a shared edge, it is not really a goal of those commands to try and align the control point structures of everything. It would not really be bad to have that, but it would put an additional constraint on the surface generation mechanism so it would not be particularly easy to do, and in the fully general case of having a "trim edge" rather than a "natural edge" (as shown above), it is not really possible at all.
Having this structure of "underlying surfaces" and "trim edges" is one of the biggest differences between NURBS modeling and polygon modeling. In polygon modeling the edges you see are always the natural edges of the polygon surface, they are one and the same. With NURBS that is not necessarily the case - this is also the reason why in the general case you can't just turn on points for edges and yank their points around (again, see that FAQ link for some illustrations on this).
But at the same time this "underlying surface" mechanism in NURBS is what makes NURBS objects work far, far better for boolean operations than polygons, because when you do a boolean on NURBS objects the underlying surfaces stay the same and only new trim curves are created, when you do booleans with polygon models the polygon surfaces become fragmented into a zillion little separate surface slivers and become harder and harder to manage with each cut.
Anyway, hope this helps shed some light on this!
- Michael
