Regression testing conundrum

Our company wants to do regression testing across builds. The way we update our software, there's no possibility of rollback and there's no way to have two versions running on the same system.

So the problem is, we only have one shot to do a regression test, unless I set up a disk image and return it to a certain state. This doesn't seem optimal.

I was thinking of having two virtual machines loaded on a single machine, one with the current build, and the other with the next build. This way I can test across builds multiple times and not just once.

Does anyone else do it this way, or is there a better direction?