Thanks to everyone who attended the "Virtualisation Live & Uncut" TechNet evening at the Microsoft office in Reading last night - I had a fantastic time, and it was great to have such an interactive audience. I think I now understand a lot more about your concerns and areas to focus on. The unscheduled follow-on "unplugged" bit was great - if I repeat the event, I'll bring a guitar with me. So, in that vain, if you missed yesterdays event but would like a rerun, please add a comment to this entry - if there's enough interest (maybe in Birmingham/Manchester?), I have the means to get it organised.
So, back to a question asked of me, and also one which I've asked myself but never given an answer serious consideration. Suppose you use differencing disks to build say 20 virtual machines all from a single base image. You run them for months for testing purposes, but discover that the size of the differencing disk is growing huge - to the extent that you would have been better off (in terms of physical disk utilisation) using dynamically expanding or perhaps even fixed disks. One downside of a differencing disk is that there is no means to compact them as you can for a dynamically expanding disk - it just grows and grows.
A solution, although not perhaps the most trivial to implement, is to use VSMT to perform a V2V (Virtual to Virtual) migration, deploying the image back to a dynamically expanding or fixed disk. Not ideal, but exactly something I'll be doing this weekend (but only 5 machines in my case).