If you have several computers tasked with doing different regions of an image (which is what network rendering does), if one machine can calculate 50 samples in 2 mins, and another lesser powered machine only gets 10 samples. You now have a final image with 'squares' that look less done than others. It's a restriction for your own good most likely, so you don't waste time on an image that will ultimately look bad in the end.
You should still be able to use Max Samples though, they use the same 'real-time' engine, but instead of terminating the calculation based on time, they do it based on # of samples (obviously), that way, less or more power only correlates to less or more time per region, not less or more quality.