I've asked the experts several times and the closest answer I've received is that the Arcade Card ram is too slow for anything but simple segment storage.
The Arcade Card memory isn't randomly accessible ... you set a starting address then grab a byte at a time.
Think of it like a modern SSD, it is fast storage, but you need to load things into real work RAM if you want to execute code. (There are limited exceptions, but the basic point is accurate.)
OTOH, you can pretty much copy sprites/tiles from the Arcade Card to VRAM very fast without copying them to work RAM first.
That's what it was designed for ... to give CD games access to a lot of sprite and background data, just like the SFII cartridge.
I believe that the SuperGrafx was given so much RAM as overkill to compensate for lazy developers who couldn't push the hardware well enough through skill and work and to just make it future proof in general.
I'd suspect that it was more to allow for developers to store compressed data on the HuCard, just like people do on the SNES and Genesis.
You need to decompress the data into RAM before you use it, and the original 8KB in the PCE is just too small to store much sprite/map data.
"Yes", you
can decompress directly into VRAM, and that's great for some things, but it's not really useful for "real-time" graphics, and decompressing into RAM gives a programmer a lot more options.
I'd love for one of the programming experts who have examined CD games to say whether any CD games definitely have or have not used RAM for more than content.
Well, I've only looked at one PCE CD so far, but it's doing some pretty sophisticated dynamic asset juggling in-and-out of RAM.
And by "assets", I'm including "processor-code" and "scripting-language" as well as graphics data.
Because too many console war fanboys like to say that the CD-ROM is a major hardware upgrade and that the System cards don't allow larger content segments, it only upgraded the Work RAM so that the PCE could handle running 16-bit quality games.
Well, that's because way too many "fanboys" are talking out of their behinds because they've never written a game!
It's so easy to poke fun at SNES owners ... just ask them why their 8-bit processor runs at 1/2 or 1/3 the clock speed of the PCE's processor.
They'll always come back about how much better the SNES video chip is, and there's a good argument there ... but Hudson and Nintendo both had access to similar technology back-in-the-day.
Nintendo decided to use the complete VRAM bandwidth to produce more screen layers. That meant that you had very little time available per-frame for the CPU to actually update any graphics.
Hudson only used 50% of the VRAM bandwidth to produce the PCE's background and sprites, and let the CPU have free access to video memory at any time.
As a programmer ... I much prefer Hudson's design choice.