Conclusion
The residuals above indicate that the Z values are unreliable without GCPs.
The X and Y values are better than the Z values which are close enough for an non-rectified orthomosaic if no horizontal measurements are needed to key features in the ortho.
I am a little disappointed in the RTK results with the M600. I have read and been told that no GCPs are needed with RTK. Each image EXIF tag might be more accurate to adjacent images however when the images are aligned there is still distortion without GCPs.
Without GCPs to a known datum, the processing software block bundle and least squares adjustment has no fixed position to constrain the model to.
Most of the mapping that I do is for 3D design level for engineers. The results have to be within ASPRS accuracy standards and constrained to a known datum.
Does RTK help with the accuracies and eliminate the need for GCPs? I am not convinced that they do, at least for 3D design level mapping.
For GIS Mapping RTK would be useful for results to a specific projection.
Is RTK worth the $5k+ price tag, you will have to be the judge on that. I don't think it is worth it. The RTK unit for the M600 sits in a box on my desk. I don't use it.
It really boils down to constraining the model to known positions (GCPs) on the ground to get the most accurate results.