The existing wording is simple and generic, it suggests using memory log on the main level, to walk through ngx_cycle->(next->)log.
I have doubts that documenting aspects of using it on sublevels (http, server, ..) may be as useful, given that memory log "does not have significant impact on performance even under high load" (c).
Further, in addition to implications with finding an appropriate log reference, using it that way may not catch all possible cases, e.g. if crash happened not in a request context.
That said, although possible in practice, it'd rather refrain from documenting this.