In my experience this practice isn’t obfuscation or for performance. The whole point of array processing languages is to amortize any fixed operation cost over the span of the array.
The notation-as-a-tool-of-thought camp prefers short identifiers on the principle that the more concise the expression, the easier it is to comprehend in total. This is not at all dissimilar from the general practices of mathematics.
Once you get used to it, it's just faster and stop noticing the 'oh line noise' thing. But like learning a natural language, you need to get to the point where you stop translating in your head and you think in the operators. Once you reach that, it is very hard to go back.
Part of those two days of commenting was that since K is interpreted all of the developers would hand-obfuscate all of the code so that the variables were a, b, c, etc. so that they would be slightly faster to parse than a multi-character string. This sort of thing is done all the time now, with JS to make it load faster, but his team at least didn't have any scripts to automatically turn developer-friendly code into interpreter-friendly code, they worked with the interpreter-friendly code version.