(2) Are your samples fused or just powder pellets?
(3) What quantification method you used?
Sometimes when you choose quantification in terms of oxides then all metals (detected) are theoretically taken as oxides which may not be the case...so the stoichiometry fails.
In XRF usually you detect from Na and above so there is an undetected dark matrix wherein all characteristic X-rays from especially low Z elements are attenuated and hence there should be a correction. The correction methods are different in different softwares. If you are using Fundamental Parameters where you consider that the sum of oxides is 100% it may not be so in the sample actually.
There should be an option to normalize your quantification to 100%...what quantification method are you using?
If the sample's metal or alloy use "Fundamental Parameters" (it should be available with most commercial EDXRF spectrometers) and DO NOT select quantify as oxides...that should solve your problem. All the best.
Oxides is just a equivalent component for elemental analysis. However, this do not mean that all elements are associated to oxigen, this is just a representative way to show the result. If you are using Fundamental Parameters, the sum of all components should be 100%. If you are using empirical calibration curve, so you can express results as metallic elements.
Oxides is frequently used in oxigen rich samples, such as geological materials, because O is the most abundant element in Earth crust (~50%). And, most of XRF device are unable to analyze it.