i think it is a mixture of performance and... yes readability.
since most of the code in axoloti firmware and objects uses fixed point math, there are a lot of conversions going on, once you get used to it those conversions are much easier to read with bit shifts then with actual divisions.
a good example would be the midi handling. in the patcher midi is represented as a blue "signal" (27bit i believe), but it comes in as a 7bit value into axoloti. you can look at the code and simply see that it is bit shifted by 20 places to the left when used in the patcher, which is very well understandable.
i remember quite a bit of discussion in the early days of axoloti about bitshifting and using fixed point math as opposed to "standard" multiplication and division and floating point. as i understood it, a couple of things (table read etc.) were much easier with fixed point, and the signal precision is also better, since it is easier to preserve the relevant bits in fixed point. and bitshifting still seems faster then mul/div.
this is at least how i remember things