It doesn't make sense, it should be increasing as the value(400, 600, etc) increases. I think you should scale those numbers up until the length of streams increases as the assumed value increases. the outputs you mentioned valid if you set the equation like this:
flowacc>400
flow acc>500( lengths of streams start decreasing, any cell that has less that 500 cells flow into will be out of my stream network)
flow acc>600. ( lengths of streams start decreasing).
Generally, the assumption of acc flows depends on the size of pixels and raster, the larger watershed you study, the higher value you need to assume.