I once got tripped up by exactly that effect. The project was pull rolls for a paper napkin folder. We did not want to pinch the paper between a pair of rolls because that could squash the emboss pattern, so we tried a pair of S-wrapped rolls similar to the diagram below.
The input roll of the folder grabbed the paper using vacuum, so could only pull with about 0.3 lbs force. The necessary tension at that point was about 3 lbs in order to keep the web taut after the embosser, and pull it over the folding plows. The tension increase is provided by pull rolls. The proof of concept test used a pair of scrounged rolls about 12" diameter covered with tread tape. It worked perfectly. The actual rolls were about 4" diameter knurled steel. The measured coefficient of friction was 1.7 to 2.0 with the tread tape, and about 0.4 with the knurled steel.
When the machine got up to speed, about 1000 FPM, the pull rolls stopped pulling. It turned out that the centrifugal force was equal to the back tension at that speed with the smaller rolls. The design was locked in by that point, so we adjusted the gap between the pull rolls to nip the paper just enough to make it work. We had hoped to eliminate an operator adjustment, but at least the machine worked.