This is quite probable. As 2 seemingly identical pipes can have slightly different readings. In some cases the ones that you would swear should ready higher will actually read lower and vice versa. I have given up on figuring out any rhyme or reason and just go at each one by trial and error until I get it dialed in.
You bring up a point that leads to a pet peeve of mine actually when it comes to lumen numbers posted in the flashlight world in general.
Anytime I see something like “light A made 846 lumens” I just laugh, there is no way in heck that a DIY sphere is anywhere close to accurate enough for that kind of reading. Some are better then others but even the best I would not put better then 10-20%.
Technically all Cree LED’s we use have a 14% tolerance from cree, meaning that no sphere can be more accurate then that. The flashlight driver/batteries ect add even more variance issues on top of that.
This is why you virtually never see exact lumen numbers from me, I always round them (usually down). I like my sphere and I think I have it reasonably well calibrated but I am under no delusions that it is still a DIY home made PVC version of a $10k sphere. These are NOT precision devices.
I would recommend and ask that people using these also round the numbers so as not to give the allusion that these are more accurate then they are. Posting exact numbers down to the single digit makes people think you can actually get a reading down to the single digit.
So back to the original question, from the comparisons I have made and looking at others (My and Djozz talked about this a few years ago) the DIY spheres you see in the community vary widely. 15-40% variance is common from sphere to sphere.
A lot of this comes down to calibration method, flashlights are simply not a remotely good way to calibrate a sphere, you have to take hundreds of readings and average them out to get something even close to a reasonable calibration. That assumes you can trust the manufactures ratings in the first place (which unless they measure each and every light, still has a 14% tolerance for the LED itself).
LED data sheets are much more consistent since Cree has a big motivation to get the numbers right but still have a 14% tolerance range built in.
So in the case of these spheres, I calibrated them with LED datasheets, this means at the very best they can be ~14% accurate to actual lumens since that is the tolerances on the LED’s themselves. The end result can not be better then the calibration method. That is simply how things work. This is why professional calibration LED’s cost thousands of dollars.
I would add a few more percent to that range as naturally there are more tolerances to take into account besides the LED’s themselves, such as heat, power supply variations ect.
Now consistency is another item altogether and that is much much better. People like to confuse consistency and accuracy. On my sphere I have a few “standard” lights that I will regularly check the output on (the same lights I am calibrating the new spheres with) and over time excluding dust build up, I get maybe a 1-3% variance at most (although I would officially rate it at ~5%). This is mostly due to the atmospheric conditions like humidity and such.
The spheres themselves seem to be coming out to well within ~3-4% on initial calibration compared to each other here. Figure that needs to be widened some once the post offices gets done with them and they are put back together but still very comparable. So the most important thing is the consistency between them as you said.
I just don’t want anyone being under the delusions that these are precision instruments. For that you will need to spend 100x the price.