Game developers will always want more of everything(except less cpu cores and more GHz), nothing is ever enough. Console maker on the other hand works within constraints of cost and needs to optimize hardware within that cost. It just might be that more memory doesn't give as good overall experience as some other configuration(i.e. fast caching+swapping). It just might be cheaper to solve part of the multitasking problem by clever programming than throwing more memory in.
I don't expect to see more than 4GB memory next gen, though I wouldn't be surprised if next gen only has 2GB.
For ram some analysis would be needed that what kind of processing power can consume whole 4GB(or more) worth of ram every frame and is such power doable next gen. On the other hand of equation is stuff like megatexture which starts from the fact that there is only so much visible in each frame and streaming the proper content in is right thing to do. Extra ram is just extra cost if the RAM is not used to do anything but cache data that will be used 2 minutes later once user is closer to the end of level.
edit. Just to make this more explicit, if console maker has design that uses let's say 2GB of ram efficiently doubling ram would imho. require roughly doubling the computing power and ram speed also. Otherwise either of the designs was imbalanced to begin with(either too fast ram for 2GB system or 4GB system will be either computing or bandwidth bound). Doubling ram from balanced hw is not just the cost of motherboard area and ram cost, you need faster ram, cpu and gpu to balance the new design. I think consoles is all about balancing the design within the given cost and other constraints(size, noise, reliability,...). On pc's it makes all the sense in the world let some users slab in more memory and used it as expensive cache, on consoles that just doesn't make any sense if you are the guy paying for the millions of launch boxes sold at loss.