Bob,

> I seem to remember some years ago using Smalltalk VSE on NT. 
> Smalltalk garbageCollect or its equivalent reported some 
> *really* large number, either gigabytes or 100's of megabytes 
> (on a machine with 32 meg of real memory).

Yes, that's one of the problems. One has to consider that memory is a
dynamic resource and may therefore change at every time. However, one might
argue that there are two pieces of information that (if made accessible)
could help determining what's going on. One is the amount of physical memory
(both, max and free) and the other one is the amount of swap space (both,
max and free). These numbers are basically 'hints' about the amount of
memory that can be allocated quickly and the amount of memory that can be
allocated but may take a while.

> When something 
> started using too much memory, the closest indication one got 
> was pronounced sluggishness and, if you listened closely, a 
> disk drive earning its paycheck.

There's an interesting question about how much of this would relate to a
system like Squeak. If you think about say Alan's demo images then you
almost always have a relatively small working set. It's basically the
current project plus a few megs of code which keeps heavy swapping basically
limited to project switches or full GCs (these *would* hurt).

> I spent two years working in 
> that environment and never got over the hopeless feeling of 
> not knowing how much memory I had been allocated and how much 
> was left.

Well, 32MB in NT "back then" was certainly not a lot but I've done it myself
(using VW2.5) and in a way I had never the feeling that it was unbearable.
Certainly better than trying to run Squeak on anything alike in which case
you couldn't even start without eating up all physical memory and a good
deal of swap space.

Cheers,
  - Andreas