Are you using 16bpp in both cases?
Ignoring the obvious size difference, the main difference between 800x600 and 1024x768 is the fact that 800 and 600 don’t relate in any way to “nice” power-of-2 numbers. It’s possible that the hardware or software has to do something ever so slightly differently to account for this, which takes it a tiny bit longer to do. The difference you’re seeing is only about 200ns, which given the testing environment (a home computer) could be considered a stastical error.
More interesting would be working out whether you’re seeing a 30% or a 200ns increase in performance. Draw a grid of 500 quads or something and see how the numbers come out then.
Edit: Given that 1024x768 is such a popular resolution, it’s possible that the drivers have been slightly tuned to reflect this. One more thing to think about!