What version of TBC are you using?
Does the line have a parabolic curve?
IF the segment being offset is only a portion of a vertical curve then the curve has to be linearized into multiple segments that approximate the original curve. Hence "random" VPI points. The elevation at the ends should be the same. From the sample I tested we are going to make a small change to increase the tolerance used so the elevations match to a higher precision.
Can you send a file with just the lines you are looking at, so we can evaluate your specific data.
thanks
tim