It's more likely that something other than oversampling is at play here, but you can test that hypothesis and probably learn something the process, or at least make some interesting observations that'll inspire further investigation.
The first challenge is objectively demonstrating what oversampling actually does. You do that by intentionally creating the one (and only) problem that oversampling solves, namely aliasing, and measure it so that you can see oversampling at work. To then determine if SONAR is correctly oversampling on export, compare a spectral analysis of that test file with and without the 2x oversampling enabled.
This isn't as complicated as it sounds. SPAN will clearly show aliasing with a test tone (but not with music) as spikes that are not multiples of the tone's frequency.
With some experimentation, I predict you will find that a) aliasing is not as common as you might think, and b) the presumed benefits of oversampling are often greatly overstated.