I do a lot of 'small data' analysis. Being able to simply load my datasets into R or Python and start playing around with them is very important. Even if my initial data set is only ~4-5 GB in memory I can quickly end up with 4 or 5 different copies in memory when testing different approaches. Obviously I can work around memory limitation by using on-disk data structures, avoiding copying data and being careful with only keeping the most important data in memory, but the less I have to think about that sort of stuff the faster and easier it is to work.

Let me add: what are you guys doing that requires 32GB off line? For 34 USD a month you can rent an i7 4770 w/ 32GB RAM, for 64 USD an i7 3930 w/ 64GB.

Might be making games. UE4 for example can eat up quite a lot of memory when doing things like building lighting and whatnot.