Further information: it seems that people are noticing the same anomaly in CM6631A/CM6632A-based cards from different manufacturers when running Linux or Mac. A commenter in this blog finds a hardware-based solution: http://archimago.blogspot.co.uk/2013/03/measurements-adaptive-aune-x1.html

He removes a hardware power-on reset chip and his Mac suddenly discovers the card's USB2.0 settings rather than the USB1.0 only that prevents use above 16 bit. A thought: do C-Media provide a reference hardware design that includes a reset chip with an unusually long reset, or something like that? Perhaps Linux and Mac are too quick off the draw, and fail to give the chips long enough to reset before interrogating them..? Something like that?

It would be great to know if anyone has read this!