That rubbed me the wrong way too. Even the hackers at MIT documented their hacks, they even wrote up a memo explaining a bunch of them so others could understand and build on them - see HAKMEM.
Documentation was much more extensive and available back then. Systems frequently came with full schematics, and you could call the design team on the phone if you wanted. DEC would get phone calls about the PDP-10's RIM10B bootloader right up until the retirement of the 36-bit processor line, and they did their best to explain its tricks.
The loader was made to fit entirely in the processor registers so it didn't touch the memory it was loading. To do this it made use of a specific and documented aspect of the processor - that being that the first thing it did when executing an instruction is to determine its effective address, and nothing the instruction can do will have any effect on its own effective address calculation.
They had two bold-print warnings about this in the processor manual, both before and after the RIM10B source code, but some people still required more explanation. For those people, the explanation was given.
In Mel's story, he grinds out a very well optimized program, and while I can appreciate the skill it takes to do that, he documented none of it. This was customer-facing code. That's unacceptable even by their standards, and even his own co-workers of the era would have thought he was an asshole. A skilled asshole, with skill worthy of respect, but an asshole nonetheless.