Actually, I have a Tanita model, and it was perfectly consistent for me.
The variance was no more than 1% at any time.
I started out with an indicated 28% and am now at 18%.
My estimate is that my actual BF% is around 22-25, but that's not the issue.