I had a similar problem. I took my wort samples from batch sparging and the wort before starting the boil, and they were always too high compared to the final boiled wort checked by my hydrometer, and compared to all my pre-brewing calculations.
This was making me crazy. I recalibrated my hydrometer in known concentrations of sugar in water, and calibrated my refractometer witt the same solutions and the two instruments were within 2% of where they should have been, and with each other, but my results were consistently off. I began to doubt my volume measurements so I calibrated by one gallon kool-aid pitcher, and it was within a couple percent too.
I finally figured out the problem. Leaving the sample (a couple ounces in an open container) to cool before the reading was letting it concentrate by evaporation of some of the water from the hot sample.
Now I put the sample in a small jar with a cap, and measure when cool, and my readings are right on the money, compared to my final hydrometer reading. I even swirl the condensation water in the upper part of the jar back into the wort.
Maybe this is what's happening to your samples as well, fast evaporation of some of the water out of the hot wort as it's put onto the refractometer lens.