I have identified some Neon defocused data taken on SM2, on 2021-01-20. I have then identified a similar data set on SM1 (September data on Subaru, taken on 2021-09-11). I have computed flux in a random 200x200 region on SM2 in the blue and red arm - and then in the identical region in SM1. (in blue arm on SM1 origin of bbox is at (40,3900) for blue and (40, 500) for red). See figure flux_red_blue which shows the regions over which I summed the flux where you can see that these are the regions that encompass the same spots. I have then computed the ratio between the blue and red areas, for both SM1 and SM2. The ratios between the blue/red arms are highly consistent between SM1 and SM2. (SM1 ratio of flux: 0.5594, SM2 ratio of flux: 0.5636)

I will do additional analysis over the region which is not pushed to such red extreme of the blue channel to avoid potential problems.

Naoyuki pointed out that B2 in particular had problems with contamination of the lens and that might have influence the result. More investigations is obviously needed.

Fabrice also pointed out that there is SM3 data available which can be analyzed as well. 

SM3: There was a level of miscommunication between Fabrice and me. He mentioned that they took continuum data on SM3, but I assumed that this data was taken on both r- and b- arms. That was not the case. The only data taken on both arms on SM3 is HgAr 1 second data, which is super hard to analyze given that the ratio between lines changes as a function of exposure time. As such, I do not think that firm conclusions can be achieved without a lot of work and care. Both Robert and I feel that this is unwarranted at this point.

SM2: I analyzed the data in a few ways. I take large sections of area (boxes at least 200 pixels wide) in both red and blue in order to catch all of the flux coming from the fiber, even if it is scattered a lot. The confusion arrives because I am getting different results on different fibers. On the fibers which I trust the most (psfs look most reasonable) I am getting the same result as in SM1. For others, I am finding that blue flux is even more suppressed. I also tried to analyze both raw and detrended images and get similar results - having said that, the detrended images in b2 shows clear amplifiers structure and residuals. I will get in touch with Romain to see how he detrened images. 
Main conclusion - I find no evidence that SM2 has more flux than SM1 in blue bands, but a final check is needed (when I also plan to add figures - see the poorly detrended b2 here). 

After some discussion with LAM team, it is not obvious that better detrend is possible, as I have used exactly the same biases and darks as they have used. As such, I will be closing this ticket with the conclusion that the results are broadly consistent with the drop between red and blue arms, but poor conditions prevent from making firm conclusions. The plan is to take data in both red and blue arms at the same time when that is possible at LAM (at most optimistic late February or March)