Attached is a file which demonstrates how the first three gradients should render (if I read the SVG standard correctly) by simply substituting the filtered gradients by gradients which mimic the filtering. I have no idea in what way, exactly, the current reference PNG is wrong, but it most definitely seems to be very wrong:
 - The first third of the "table" gradient should be a transition from yellow to black, without any green. This can be easily verified by considering that in the original gradient went from #ff0000 to #00ff00 and that the two transfer functions are mirror versions of each other.
 - At the far left of the "linear" gradient the red channel in the reference PNG reaches values above 200, while 255*(.5+.25)=191.25.
 - At x=133, the original gradient (at the top) has green value 207, the "gamma" gradient should therefore have value 255*(2*(207/255)^3+0)=272.8... (so 255, as the values are clamped). In contrast, the reference PNG has value 185.