LT3082/LT3092 500+kHz disturb causes output offset.

The LT3082 and LT3092 are the same part.  The datasheets are not identical, though.  I don't know enough about your circuit to try it myself in the lab.  My guess is that your noise occurs at a low-frequency repetition rate that is within your measurement band but the noise actually has high-frequency content that the LT3092 cannot reject.  My advice is to observe the noise in the time-domain with your o-scope and try to filter or shield the noise at its source.  Ferrites plus capacitor filters or even RC filters can be used for the filtering purpose.  The compensation capacitors you mention are just for stability and not for filtering.  My guess is that your circuit does not have a stability problem, but you can check just by doing a line transient and seeing if the current rings or is damped.

The LT3082 and LT3092 are the same part.  The datasheets are not identical, though.  I don't know enough about your circuit to try it myself in the lab.  My guess is that your noise occurs at a low-frequency repetition rate that is within your measurement band but the noise actually has high-frequency content that the LT3092 cannot reject.  My advice is to observe the noise in the time-domain with your o-scope and try to filter or shield the noise at its source.  Ferrites plus capacitor filters or even RC filters can be used for the filtering purpose.  The compensation capacitors you mention are just for stability and not for filtering.  My guess is that your circuit does not have a stability problem, but you can check just by doing a line transient and seeing if the current rings or is damped.

The LT3082 and LT3092 are the same part. The datasheets are not identical, though. 

The price neither! 

I think you're right. We got a big improvement with a LC filter. We have still a problem at its resonance, but I'm confident we can overcome it with different component value.

Ferrite would be great but 500kHz is too low.

Thank you.