Other classics: dumping the memory of an active process (check my GitHub utils repo, I have it there) and redirecting the standard output AFTER you've started a process.
Ah, dumping memory. I once took the dump-and-restart code from GNU Emacs, and planted it into GNU Make. I was then able to read a large and complex Makefile (consisting of included sub-Makefiles throughout a large tree) which took 30 seconds to process, dump the image to disk (with make --dump), and then use the dump to run incremental builds which were able to kick off build recipes almost instantly!
Other classics: dumping the memory of an active process (check my GitHub utils repo, I have it there) and redirecting the standard output AFTER you've started a process.