I could ABX Arny's old jangling keys sample in Foobar. I couldn't figure out why or what I was able to hear. I then resampled his original high rate file with a recent better resampling software, and could no longer hear a difference. Apparently was hearing the resampler.
I'd expect in this case with resampling to MP3 on those files, and back you could hear it.
One of the more interesting cases I've seen was a test by Amandine Pras. Using young listeners, all quality gear, and minimalistic recordings. Everything was concurrently recorded at both 88 and 44 rates. Testing was done 88 vs 44, and 88 vs 88 resampled to 44. A very positive result for a difference between 88 and resampled 88 to 44. Just about 50/50 with native 88 vs native 44. The resampler was the best Izotope software. I'd like to know what it was that made that discernible.