Oh my, that is exactly the kind of headaches I get when measuring light, and why I started the process of getting more grip on light measuring (which caused more headaches).
But, assuming a consistent luxreading set-up, it sounds more like a flashlight thing than a luxmeter thing. My experience with cheap luxmeters (and your Ceto, which I use as well, in my integrating sphere) is that with all their flaws of which I tried to adress one in the OP, at least they are very consistent over time: if I take a reading and do the same a month later with the same flashlight, it reads exactly the same, and for most of the drainage independant of battery level. I have no idea what is going on electronically in those things but in the sensor's point of view I can not think of a reason why it would be off just in a certain range.
EDIT: I did not test it (yet), but luxmeters apparently can have a very bad temperature consistency. As can be read in the first table of the OP, a class C luxmeter is allowed to have an error of 2%/K, and who knows about our chinese meters? I guess that 2%/K means that if the ambiant temperature is 10 degrees (Kelvin/Celsius) different (which can easily happen when measuring outside), the reading can be 20% off (my measurements were always indoors, with temperature not varying more than 2 degC, but still 4% error is big!). It may be not that bad, but we just don't know until it is tested: higher end meters come with spec.sheets with the characteristics, the chinese are not telling anything.