Looking at their browser list, I could test all Firefox, Opera, and Chrome versions, one version of IE, and the Windows versions of Safari on one Windows 7 box.
I could add to that three VMs for older versions of IE, and a Mac (maybe with a VM) for Mac Safari versions. If the main Win7 machine is also a VM, I could do this all on the Mac.
So, that's a total of 5-6 OS licenses, which would cost $750-$1000.
At $19/month for a single user, BrowserStack becomes more expensive after 2-3 years. At the small team rate of $68/year it becomes more expensive after one year. (I could host VMs on a server and have my testers RDP into it; they can't use the VMs simultaneously but with 4 testers and 5-6 VMs that's not a problem.)
To me, the pricing is right on the edge, which forces me to evaluate the effort of maintaining local VMs and browsers (an invisible cost is most organizations) vs the potential risks of running my in-development software and providing access to my development environments to a third-party over the internet. If the pricing was lower and clearly less expensive than maintaining my own VMs, it'd be a slam-dunk decision and the other factors wouldn't even be considered. (I'm thinking about this from a "how does a developer sell this to corporate management" perspective.)