I finally remembered to post this, deviation can sometimes go crazy.
Take a look at this curve, if you rebuild it to a degree 2 curve it shows no deviation, neither at degree 4, but at degree 3 it showes a 125 unit deviation! and at degree 5 it shows a sane 0.38.
Hope this helps you ironing out that bug. Deviation in rhino script has the same issues, but have not tried it here.
rebuild gives wrong deviation.3dm (24.1 KB)