Orca is a bad idea. Very few blind people use Linux because Linux desktop sucks, and Orca is always at least somewhat behind all the other screen readers. I have met a few blind people over the years who have used it for an ethical reason or because they can't afford Windows, but no one really likes it. Linux is indeed the cheapest option, but it's also almost completely ineffective in terms of making it actually work well for your users.

For starters, even getting low latency out of the Linux audio stack is a major headache, and the synth situation is abysmal. You can't even touch the config files for these yourself because if you break either--even temporarily--you now can't use the computer. Then you get into how all the graphical desktops have accessibility issues to one degree or another and how you have to use a separate screen reader for anything outside them.

What you want if you want a testing VM that actually has value is Windows and NVDA. NVDA is free in both senses and kind of the industry standard for sighted testing now. Jaws is still more popular, but this is slowly shifting in NVDA's favor. This would be because Jaws costs roughly $1000 per user. The advantage of NVDA is that you can be sure all users can have it, and if it works with NVDA then it's very likely to work with jaws without too much more work.

But sadly you can't just test with one; in the end you have to test with all of them. Things like aria are nice if used correctly, but the aria spec doesn't say much about what screen readers have to do, and no one implements 100% of it. It's very close to the situation with needing to test on multiple browsers.