That is just the mode they set the sensor at. If they wanted to read at higher fps they could make it do that. The limitations come from the ability to process the data generated while at the same time doing whatever else the camera needs to do. Obviously you don't create a spec for the sensor that your processor can't meet. The two are designed around each other, and if the sensor is repackaged to be sold to third parties, those third parties get whatever specs were imposed on the design by the processors Sony had available.
The absolute speed the sensor can be read is kind of irrelevant if the processor has to do other things at the same time, which is what is happening with the 1DXIII. The sensor clearly can be read faster, since it does exactly that at 60fps, but not when the processor is doing AF at the same time (AF has a heavy computational load associated with it). AF processing has to be done constantly, with other data being read in-between, it can't just stop while the sensor is being read, if it did then AF response times would become much longer and you would get sluggish AF performance. So, a sensor frame read would go something like data...AF....data...AF.....data....AF etc, they are interleaved. Those spaces in-between during which AF functions are going on is why rolling shutter is 30+ ms. The frame read data is being spread out to allow AF functions to happen in timely matter instead of coming all at once. Basically AF gets priority over the sensor read. if you had fewer AF sites that needed to be polled and data processed, then you would have lower RS with the same sensor. The problem for Canon is that DPAF has a bucket load of AF sites that require attention, hence the RS, while other manufacturers are using conventional PD sites which are far fewer in number. Fewer PD sites = less AF processing overhead = less time needed to complete the frame read = lower RS.
Not rocket science.