Randomness is always important to some degree. The basic question is what is the unit of randomness: Does each character get its own RNG? A group? An instance? A region? A server? A server cluster? Even if the scope that a random number generator covers is broad, if you have a fairly simple algorithm like LCG there's a potential for exploit. Well, there's always a potential for exploit no matter what you use, but choosing well can minimise that risk a lot. (Of course an individual does not get their own RNG, but you get the idea.)
You're thinking combat table and it's true that there's not much you can do there, but random covers a lot more than that. For example, suppose I go into an instance with a scheme where I kill a mob every X seconds without fail. If I can find X such that I will always get a green from that mob, or even if I get a green item from every third mob, it's all good. The scope and randomness of the generator dictate how that works. If the RNG is on the instance scope then that's a lot easier than on a server scope for sure, but it's not necessarily impossible to do on a larger scale, particularly if you stumble across a code flaw that you can use to facilitate or if you can consistently apply some sort of frequency analysis to the results you get. If you can do that, maybe you can massage your scheme to now get "random" epics. Even if you can only pull it off once a week, it's worth it.
Even better, if you can manage to approximate the serial correlation of an RNG you can start trying predictive stuff based on estimates of consumption of the sequence. This is particularly true of LCG since the period of its low-order bits is very short. If you can find a cycle period, maybe at off-peak server times when demand is lower, you can use that. Again, you don't need to get it right every time to make it worthwhile.