It boils down to what colours do you currently have and what do you want to do with them.
Skin, for example, pretty much fits within sRGB. If you have say 99% of your colours within sRGB and want to play with them and you're not really worried about that 1% that's outside and are happy to bring it into sRGB, then doing so means you have more levels between the extremes of any given colour which means smoother transitions/gradations etc. This can make editing easier/better (subjective obviously).
It doesn't matter if future printers might expand current gamuts, because we already have printers that can exceed Adobe RGB in places and complete encompass and exceed sRGB but if your colours don't exist outside of sRGB then extended printer gamuts are not an issue - you might as well make sure you have the best progression of levels in the colours you have.
If, however, you have say skin tones (sitting virtually entirely within sRGB) but a large amount of say blues and reds sitting outside sRGB and even outside Adobe RGB then you wouldn't want to compress your working space if you wanted to keep those colours relative (or as relative as possible given your output gamut) to the skin tones.
There's nothing inherently wrong in working in a large colour space, but you *might* find an advantage through working in smaller space to give you more steps/levels/tones/gradations/etc during editing.
Yes, 16 bit is cool, but open up your image and go into a levels adjustment. You have 256 (0-255) discreet levels you can adjust and if you go to curves, it's the same. That covers the entire available range so if that range is larger, the steps must be bigger between each other.
Want to use the water analogy? Get a really large bucket so that only 1/32 of an inch of water covers the bottom (1/32" deep) and another that is say 4" deep with the same volume of water. If you splash some of the water from the 1/32" deep bucket with a finger and it goes out of the bucket, you may or may not have water run in to cover the dry spot. The bucket may not be perfectly flat and the water tension might be enough that it just starts to pool, leaving dry spots all over.
To do that with the 4" deep bucket you'd have to throw out a LOT of water - almost all of it until it gets down to being 1/32". Of course that's an extreme example and please don't anyone come and tell me that water doesn't bring in surface tension effects until 1/64" or something, 'cause that's not the point.
The point is, if you spread yourself too thin you may end up with areas with no water - no data.
It's not that every single image should be plotted against all available colour spaces and then grab the one that just fits - but that *might* be appropriate depending on the image and what you intend to do with it.
EDIT: For the record, I use ProPhoto almost exclusively, but I don't process large volumes of images so if I was having problems then I'd consider going to a smaller space.
Schewe posted while I was and did a better job of explaining things anyway :-) My only contention with what Jeff posted is that in the future Pro Photo might *not* be proof against clipping from output spaces since it doesn't encompass all of human standard viewer colours (let alone the possible colours of tetrachromats, but that's a whole other discussion and pretty damn interesting :-). However, he is probably going to be right for a significant part of my life time so close enough!