This question only makes sense in Simulink.
In Matlab, you cannot tell the difference between generating the random data at 20 MHz and consuming it as it generated, vs generating the random data at (say) 1 GHz for a burst and consuming it at 20 MHz until more samples are needed.
It is true, though, that in Matlab if you had a D/A interface and A/D interface, and the generated data had to include some feedback from the incoming signal, then the burst length of generating random samples would start to become important. In such a case, though, you would be better off having the data generation triggered by the callback of the data reception function rather than trying to fix a particular generation rate.
