It could be an error in dilution, or it could be caused by a variety of integration errors. The shape of the peak may change, and fail to integrate the same way. At very low levels, the baseline really affects the area, if you zoom in repeatedly, you'll eventually see the baseline noise and how much of a small peak can be gained or lost. The optical system may not be linear over a very wide range -- although that is rare. You initial large peak may have saturated the detector ... so that it was really bigger than it appeared. A really small peak may be under represented because there is a time factor ... the detector can only look at the flow cell so many times a second, it may miss some of a very narrow peak. Remember, absorbance has to be above some threshold greater than baseline in order for the software to say "peak starts here." A low absorbance means it ignores more of the peak, so it may misrepresent the area.
Suggestion: try running some intermediate dilutions, to see if there's a sudden point of deviation, or general, allover the place weirdness.