Just like the title says. Is this due to roundoff?
I'm not sure how to illustrate it with a simpler example, which shows how little I know about the issue. I can say when I plot a function (takes a lot of code to construct) one plot has max of 50 and the other has max 6000. Do you want me to post the code or pictures so you can see?Can't you illustrate it with a simpler example?
Its well known in computerdom that some numbers can't be represented properly in floating pt format so if Mathematica does a numerical computation with them them then computational error will creep into the calculation.
As an example, if you had some expression like ##3.0*sin(x)*(1.0/3.0)## and Mathematica symbolically reduces it to sin(x) then that result might be different from ##(3.0*sin(x)) / 3.0## where Mathematica didn't see the ONES identity and did the numerical computations of 3.0 * sin(x) then dividing by 3.0.
For larger numbers the result difference might be more pronounced as floating pt numbers kep to a certain limited digit precision.