They’re voltage dividers in isolation, but in circuit they end up being in series/parallel with the summing resistors, with the series bit adding to the output impedance of the previous stage, and the parallel bit lowering the input impedance of the mixer (e.g. with 100k everywhere and the pot in the middle, you end up with 50k in series and 50k in parallel). The result is a voltage drop and a non-linear pot response (well, more non-linear than when pots < summing resistors).
Since all inputs are the same this doesn’t matter much in practice, the only difference is that the pot positions will differ a bit from the original, which is easy to compensate for (or you can tweak the gain later on). Alternatively, you can make the R3-R6 input resistors and the R7 feedback one larger to preserve the pot/resistor proportions, but for truly independent attenuation you need individual input buffers. Which is probably overkill.
EDIT: Too lazy to do the maths, but did some quick checks in a simulator, and with 50k/120k and the pots in the middle you get ~90% of the signal out, with 100k/100k you get 80%.
EDIT 2: Did a bit of plotting too. Straight line is ideal linear attenuator response, the other two 50k/120k and 100k/100k.
