My experience has always been that "not supported" means anywhere from "physically impossible" (e.g. because there just isn't that many address bits) to "we won't help you because we haven't tested it".
Some quick searching shows that others have managed to get that CPU to use 32GB of RAM:
I thought he would go as far as patching the BIOS itself, which would make it a "permanent" fix. In fact one of the projects I haven't gotten around to finishing is to patch the memory init code for an old Atom processor embedded motherboard to make it recognise more combinations of RAM modules; analysis of the BIOS and leaked sources shows that it was stupidly written as "if(512MB in slot 1 and 512MB in slot 2) ... else if(1GB in slot 1 and empty slot 2) else if(1GB in slot 1 and 1GB in slot 2) ...", when the memory controller can actually be set up generally to work with many more different configurations.
One end of that range should include 'you already have it but we're not telling you'.
I noticed the type numbers of one of my computers had RAM that was 2x the quantity advertised and indeed with some fiddling I managed to unlock the second half.
I went to the local store to buy 4MB (4x1) for a computer I was building but they accidentally gave me 16MB (4x4) for the same price. Then they went out of business, turns out it was a money laundering scam; guess they didn't know or care exactly what items went out the door.
That's bizarre. RAM is relatively expensive and has long been a substantial cost of the computer. Not like CPU binning where it really can be cheaper to sell you the faster processor and lock it to behave like the cheaper one.
I vaguely remember some Sinclair computers doing stuff like that because they used cheap RAM chips that were ensured to be error-free on the entire memory-space (so basically binning) so they used the "good" half, but a lot of the time the "bad" half was also fine?
Your definition's right in the sense of what "not supported" allows the user to do. But the term also means "if you try this and it doesn't work, we're not responsible," and that's the key thing.
Some companies (and programs, and programming languages, and so on) make it easy to do unsupported things. Some make it hard. None will accept liability for it.
Hardware manufacturers are on the hook for so many potential problems, defects, misconfigurations and misuses that it makes sense from their perspective to support actively only those things they can guarantee will always work. Not things which the hardware can theoretically do but which might not always be the best idea, things that aren't possible now but might be in ten years, etc.
Some quick searching shows that others have managed to get that CPU to use 32GB of RAM:
https://forums.overclockers.com.au/threads/p55-chipset-i5-75...
I thought he would go as far as patching the BIOS itself, which would make it a "permanent" fix. In fact one of the projects I haven't gotten around to finishing is to patch the memory init code for an old Atom processor embedded motherboard to make it recognise more combinations of RAM modules; analysis of the BIOS and leaked sources shows that it was stupidly written as "if(512MB in slot 1 and 512MB in slot 2) ... else if(1GB in slot 1 and empty slot 2) else if(1GB in slot 1 and 1GB in slot 2) ...", when the memory controller can actually be set up generally to work with many more different configurations.