Changes in sample morphology could definitely be seen in S02, but I would think the main effect of a change in sample morphology for transmission data would be a change in data quality: glitches and so on in chi(k). So: Does the data quality get worse as you change the charging? How well harrmonic rejection and sample size/alignment can have an impact too, but I would expect data quality to change dramatically if you're seeing a systematic change in S02 that was due to changes in sample pinholes, etc.
As Scott and Bruce said, S02 is only as good as the normalization, so you should be consistent. But, beyond that, why would changes in EXAFS amplitude be assigned to a changing S02 instead of a changing coordination number? Isn't that the simpler explanation?
It seems silly to try to test how well S02 can be transferred between spectra for spectra that are known to have different morphologies and local structures that are quite possibly changing in unknown ways (that is, on actual samples). Do that on well characterized standards. As far as I know, every time this as been done, S02 is found to be stable to better than 10%. Maybe there are cases where it is worse than that....
--Matt