Nothing concrete from my end, although quite a few people claim to have tested this fully, including a clanhosts representative who frequents the forums. I've tried to find a good post he made on the subject but my google-fu is weak this morning.
Understandable, so no worries there.
The issue is also confused by memory leaks. Presuming there are no memory leaks you're going to have a reasonably set amount of RAM that you need. Once you have more than that any extra will have no benefit. This brings the limiting factor to another part of the system, which will be the CPU, thanks to MC being largely single threaded.
In regards to memory leaks, more memory should, depending on the type of memory leak, give you longer stable runtime between restarts, no?
And disregarding memory (either through no memory leaks, or assuming that memory leaks are going to destroy your memory no matter way), I'd think that disk speed would be the next major limiting factor (within reason).
That your server is using a fraction of it's total CPU power only shows that it's not executing MC code across all cores (and hyperthreads) very well, if at all.
There is multithreading (for example, ComputerCraft's code, which runs in its own thread I believe), but yes, the server code itself will run itself in a main thread. And what that means is that I'm only seeing around eight to sixteen percent of one of my four (and four virtual) cores, which is still an extremely small fraction of usage compared to how much memory it demands (memory leaks or not, because the vast majority of memory leaks are not enduser addressable).
A more interesting metric is finding the core MC has it's main thread on and seeing that usage in isolation.
I normally set the affinity of the server to core five, which usually bounces between five and twenty five percent usage. It will sometimes spike to a bit under fifty (likely when ForgeBackups hits).
RAM is very important, don't get me wrong. If you go under the amount you need it's obviously a horrible experience. My point is that beyond that, chucking more RAM at a server is not going to make it faster.
Yes, that is an excellent point: RAM only makes things run faster when you are satisfying parts of the saturation range. When, instead, you have passed the range of saturation (commonly given as 256mb to 1024mb per player), more RAM will not do anything useful. That is a very good point to underline for everyone involved, as it is one of the main things people often mistakenly think.
