That's exactly what he was doing. The codebase had reached the point where it was overly complex, and he wanted to simplify the design. If look at his conclusion he only refers to making design decisions, not writing new code or designing algorithms.
[Insert obligatory joke about Linux, BSD, Perl, Lisp, C, etc here...]
Admittedly I've never written a compiler in 360 assembly on or off LSD, but it seems like the better thing to do would be to take a step back and think through the design better (that would have been my first reaction at least...).
Put another way: I'd say LSD is code smell. :)