> Anyway, I’d remind everyone that “using” RAM doesn’t mean “would not function with less RAM.”

Except when something really does need more RAM, and fails. LLVM for example having, somehow, become a bit chonky and now fails to compile on 32-bit OpenBSD systems because it wants more memory than is available. Less bloated software of course does not suffer from this problem, and continues to run on still functional 32-bit systems.

> Many applications just use a lot if it’s available.

Xorg is using 92M, irssi 21M (bloated, but I've been lazy about finding something leaner), xenodm 12M. That's the top three. Oh, Windows? Yeah. About that. Best you can hope for is not to catch too much of the splatter. (What passes for Mac OS X these days also seems fairly dismal.)

> RAM is not really something you explicitly ration.

Paperclips were hung on the rack doors to make it easier to poke the wee little red reset button when some poorly written software went all gibblesquik (as poorly written software is wont to do) and the OOM killer could not cope and, whelp, reset time. Elsewhere, RAM is explicitly rationed—perhaps certain aspects of timesharing have been somewhat forgotten in this benighted era of bloat?—and malloc will return NULL, something certain programmers often fail to check for, which is generally followed by the kernel taking the error-ridden code out back and shooting it.