Theoretically the lowest noise figure obtainable from any device is limited by any components in the signal chain with the highest thermal noise. The first component in the chain would be the detector circuit and on a Universal LNB this would be a pin diode At Ku band the detectors are rated at manufacture to about 40K which converts to a figure of 0.5 dB.
Presently the LNB market is active by those selling what appears to be an extraordinarily good device (and good luck to them).
Some use the best of the component batch - and end up with good performance overall (rare)
Some modify existing LNBs by the use of fancy smoothing circuits to eliminate any further incoming noise from the power supply - there are definite improvements when used with cheaper receivers, especially those with switch mode supplies (few )
Some simply take the lowest noise figure during tests at a particular frequency, and then claim that this is the figure of the LNB (so you could have a 0.5dB LNB at 10.8GHz which has a 1.1dB noise figure at 11.6GHz, but it will be rated at 0.5 (depressingly common)
Most however will blatantly lie about the noise figure to increase sales, and to hell with the consequences
Rolf is right that the better solution is to look at other components in the system, starting with the dish and its alignment
