Sorry to quibble, but this isn't entirely correct.
Even a 'perfect' LNB will exhibit thermal noise, which is easily calculated. This represents a lower bound on what can be achieved in the real world. A real world LNB will always produce more noise than this amount. The usual definition is normally said a little differently, but a Noise Figure (NF) rating provided for Ku LNBs is the difference (in dB) between the actual noise power of the LNB and the theoretical limit of noise power. C-band LNBs are usually specified in noise temperature, which is just another way of stating this.
I would tend to agree the NF ratings of most LNBs are dreams at best and lies at worst. In fairness, for practical reasons such specifications are usually provided as nominal values for a large population of units, rather than a maximum value that will never be exceeded. A few days ago I published a short piece here on how much variation there can be across the frequency band. I've toyed with doing a more explicit test that would plot NF vs. frequency and compare this to the individual test curves provided with some LNBs. That could be quite ugly.