The point isn’t to demonstrate maximum error. It’s to see if the software could get results that fall within the allowed margin of error. This was only possible through a mesh size of 5 mm.
This isn’t ‘real engineering work’. It’s a standard that has to be followed if we were to say that PrePoMax/CalculiX is validated against ISO 10211-2017. Plus when you want to use a software for engineering/academic work, it has to be validated against some standard. COMSOL, HEAT2&3, Mecway, SimScale, QuickField, FEMAP, and WUFI all use ISO 10211.
You should aim for a scientific balance between accuracy and time/resources. Saying “I don’t have time” isn’t good practice, and a lot of engineering catastrophes happened because people were trying to save time/resources