A good starting rule-of-thumb for absolute detection limits is 1 ppm by atomic weight. You might be able to do better sometimes, but there are situations where XANES at 10 ppm is very hard. For sure, a matrix of light elements is much better than a matrix of heavy elements.
For very dilute samples, one will be using fluorescence XAFS measurements with a solid-state detector or know very well why you are doing something different. These solid-state detectors and electronics are fundamentally limited to have energy resolutions of ~120 eV (often 250 eV) and maximum total count rates of 5 MHz (often 0.5 MHz). Many beamlines use "a handful" (2 to 16) parallel detectors, and some have up to 100 (but often with each having a lower individual maximum count rate, and perhaps less-than-ideal energy resolution).
With a count rate of a few MHz total and a sample with 1ppm of "element of interest", the elastic and Compton scattering and/or fluorescence from other elements will dominate that total count rate and the energy resolution will give non-zero background in the fluorescence spectrum. That means that even seeing a peak from 1 ppm of an element in an X-ray fluorescence spectrum with a solid-state detector is challenging. Not impossible, but definitely not routine.
For sure, adding more detectors or counting for a long time can help. But those are linear in time and the number of detectors (and no beamline has 1000 parallel detectors). Low Z matrices like water, biological material, carbon-rich materials are easier. Samples with nearby or overlapping fluorescence lines are much harder. That is 10 ppm Zn in water: yes, 1 ppm Zn in water: maybe, 10 ppm Zn in CaCO3: maybe, 100 ppm Zn in Cu metal: no. For sure, XANES at 1ppm is sometimes possible. Getting interpretable XAFS would take a lot longer, perhaps days of counting.
Using filters and/or Bent Laue Analyzers in front of a solid-state (or integrating) detector can sometimes help to eliminate the unwanted scatter signals before they get to the solid-state detector. Using crystal analyzers ("wavelength" vs "energy" dispersive fluorescence) can help - they have lower backgrounds and are not limited by the total scatter - but the solid angle for these tend to be small. Using crystal analyzer arrays are probably really needed to get the best detection limits. A few beamlines do regularly do HERFD analysis with arrays of crystal analyzers, and many of the rest of us are trying to catch up. Still, I believe that "1 ppm" is around the state of the art, if not "heroic".
All of that is for the detection limit of an atomic species. If you are asking about detecting Cu in/on/with Pd nanoparticles with CuO also present, the answer is far worse. Cu XAS measurements will be an average of all Cu atoms in the illuminated volume -- you cannot avoid the CuO. Seeing that 1% of the Cu atoms are bound to Pd and not to oxygen would be very challenging.
Hope that helps,
--Matt