Do you think the image of SD14(SD15/DP1/2) is cleaner than that of SD1m(DP1m/2m) ?
I don't know about Merrill's but SD10's and, according to Arvo, SD14's have a separate saturation value for each channel embedded in the metadata of the X3F file. Approx 6500 on my SD10. BUT, if you deliberately saturate the sensor then the X3F data can be up around 10000! So, there's quite a bit of headroom over what Sigma considers "saturated". Therefore, exposure beyond "right" is feasible but presumably gives non-linear highlights and a higher possibility of individual channel clipping.
In one of their early papers, F7 sensor I think, Foveon said that the well capacity is 77,000-e but 44,000-e is preferred for better linearity - a clue perhaps?