But it is information that IS needed in order to make an informed judgement, if you do not have such information it is just guessing what is happening on somebody else's system.
At high cache sizes, a crash IS pretty much inevitable, almost guaranteed to occur, BUT, does it also happen ON THE SAME machines without Gibibyte cache sizes??
You don't have that data and you will not be able set up the same conditions to ascertain the results, so you have to get others to set and test different conditions, qBT is never going to be run solely in an ideal environment, with perfect settings so you need to know the "safe working limits". Currently it is patently obvious that a maximum of 1800 is WAY over the safe point, but what is a 'safe' maximum that WILL prevent users from setting values that they think or have 'read somewhere' are useful but are actually detrimental to performance??
The only people who can help determine that, ARE the ones that are having the problems and are willing to help themselves and the overall project at the same time.
Someone commented that findings were useless because my "config was low", now
THAT shows a lack of understanding of BitTorrent in general, my configuration is ideal for how I run my systems. There is no "one size fits all", so for whomever that was, ... ... it's not the numbers that matter;
It is the relationship of the numbers, the ratio of connections to cache (or memory use) is very likely to be similar across configurations and use cases. The numbers may vary widely but you will only find the 'best fit' for your system by trying many different 'tweaks'.
because at those limits you can discern if the RAM usage belongs to the cache or to the rest of the program...
Should that be "cannot discern"? If so, then sure, just as you can't say categorically if it IS the huge reserved cache size or the application image that causes the crash, it takes testing at many combinations of cache size, as the only easily adjustable user variable to separate cause from effect.