Caveat: I've not used a DDR3 rig yet, so this is based on conjecture as much as anything else.
DDR3 is faster for bursts of activity but has higher latency than DDR2, so while it should be faster for most tasks it might not be for all and the latency may reduce the resulting performance boost. So unless the price difference in the boards is really small, I'd stick with the DDR2 and reuse the existing RAM. In fact I'd probably stick with DDR2 anyway. You can always upgrade the motherboard and RAM at a later date.
For your "light gaming" it will make no noticeable difference. My DDR2 based machine plays Crysis at 1920x1080 with everything in high quality with only one place in the game where it even stuttered (and I think the graphics card was the bottleneck there, not the rest of the system). Having DDR3 RAM on your graphics card will make more of a difference in gaming than the better RAM on your main board will - the tasks GPUs tend to perform benefit more from it than current generic CPUs will and with modern games the graphics card is more likely to be the key bottleneck rather than your CPU or RAM.
For VMS I'd wager that the overhead of context switches between VMs and other virtualisation overheads will pretty much dwarf any difference you get by the DDR3 memory being able to send them blocks of data faster.
For development: unless you are doing serious number crunching on large datasets (if all your inner loops work on small datasets, they'll be sat in cache the whole time anyway), you are not going to see much difference. While visual studio is fairly heavy it is not going to overly tax a machine enough to see the difference, database work too - your CPU will spend more time waiting for disc IO then it will waiting for memory reads.
Also remember that modern CPUs come with large L2 caches built on. Most tight, and many not so tight, loops will be served by that cache rather than needing much by way reads from main memory - and if you have many processes active at once to the point where all your L2 cache is getting clobbered each time around the scheduling queue then the overhead of context switches is going to dwarf any difference in bulk RAM response (and that higher latency is going to kick in too).
When I bumped my CPU up a couple of notches some months ago I kept my existing AM2+ board for the above reasoning, and haven't had any regrets saving the cash. Well, when I say "saving the cash" I mean "spending the cash on other stuff like a better graphics card and an extra hard-drive". DDR3 is currently, IMO, for people who don't have to think about spending the extra, especially people who are doing a fresh build and don't have to consider the economy or reusing old RAM, hardcore gamer types who must have the latest+greatest+fastest everything and believe every bit of marketing hype (they are like audiophiles at times), or people doing specialist computational work that needs the extra bulk transfer speeds between RAM and CPU.
