小中大Is the sputtering rate you used is like another question you post @ 5.01nm/min?
In this case, each 20s sputter roughly maybe removing 1.7nm. And XPS detection depth roughly is ~5-10nm say as average. So you are always detecting the XPS data in "a range".
Ra roughness level can easily be in um range. But from your description above, anyway XPS is always giving you the real Physics and your 2nd assumption seems more possible too (less Sn on particular area on sample).
If you really want to study about roughness effect, you need to actually analyze your sample to know its real roughness and then study it with XPS for different roughness how the differences would be on your samples then.
I think any direct simple "guess" from what you have now seems not very appropriate like you said.