What "Gworlds" is? Eh. Creative googling will sort of explain it:
http://www.mactech.com/articles/develop/issue_08/095-098_Guest_column.html
http://developer.apple.com/legacy/mac/library/documentation/mac/QuickDraw/QuickDraw-2.html
... (Look at the chapter about "offscreen graphics worlds) and they're mentioned in passing in the Wikipedia article on Quickdraw. To oversimplify, it basically seems to boil down to being a form of "double-buffering" in which pointers are set so that Quickdraw, rather than drawing directly to video RAM, renders a complex image element to a RAM buffer which is copied to the display device when it's finished. Apparently it was possible with these Nubus cards with SIMM slots to set the render pointers to the RAM cache on the card so the CPU would render to the offscreen buffer there instead of system RAM, and when the image was ready to display it could be simply blitted by the video card's hardware into VRAM rather than transferred over Nubus. (Which would be comparatively slow compared to a memory-to-memory transfer on the card itself.)
An application had to specifically support doing this, so having that memory in most cases doesn't make anything faster. And apparently OS updates had a bad habit of breaking apps that did do it.
http://www.mactech.com/articles/develop/issue_08/095-098_Guest_column.html
http://developer.apple.com/legacy/mac/library/documentation/mac/QuickDraw/QuickDraw-2.html
... (Look at the chapter about "offscreen graphics worlds) and they're mentioned in passing in the Wikipedia article on Quickdraw. To oversimplify, it basically seems to boil down to being a form of "double-buffering" in which pointers are set so that Quickdraw, rather than drawing directly to video RAM, renders a complex image element to a RAM buffer which is copied to the display device when it's finished. Apparently it was possible with these Nubus cards with SIMM slots to set the render pointers to the RAM cache on the card so the CPU would render to the offscreen buffer there instead of system RAM, and when the image was ready to display it could be simply blitted by the video card's hardware into VRAM rather than transferred over Nubus. (Which would be comparatively slow compared to a memory-to-memory transfer on the card itself.)
An application had to specifically support doing this, so having that memory in most cases doesn't make anything faster. And apparently OS updates had a bad habit of breaking apps that did do it.