If the test environment has 2 app, 2 web, and 1 DB server (equal to prod), then I'd try running 1000 VUs on it. If that works, then you have a pretty good idea that prod will work.
If it fails, then I would try 500 VUs if and only if the following is true;
- DB server utilization is very low,
- No adverse blocking seen on DB server
- DB server IO rates relatively low vs capacity
- No communication between the app servers
- This is a relatively low risk application deployment (neither you or your company will be materially effected if this instance performs poorly)
The main risk in this approach is that a resource other than the web or app servers becomes the primary bottleneck above 500 VUs. Most likely bottlenecks, should they occur, would be at the DB server or comm gear (routers, switches, load balancers etc).
Do you have the ability to instrument your test environment so that you can determine areas of concern? On a smaller environment, you'll be able to identify these areas with fewer users, which might not be possible with your current VU count on a larger system. In fact, you can even scale your test env down to one server each to make the bottlenecks even more apparent.