To me the algorithms for deciding whether a URL is cachable or not are, what you call it, black magic? or voodoo?
How do you determine that an object should have been cached?
Try this maybe: http://www.ircache.net/cgi-bin/cacheability.py
My conclusion, when I got intrigued by this was, I hope LTM knows what it doing... and worry only when it caches something it turns out should not, like the user settings or the wrong user or something.... Which is probably the reason why caches are very careful and choose to err on the safe side.