The phone's callstack takes a lot of special actions when calling 911, including locking onto any tower even if it is not in the preferred roaming list. As with many exceptional conditions, these code paths are often not as well tested.
If you need to do automated tests, there is a test flag in the android dialer API which will make the call but mark it with some test flag so you get a recorded message not a human answering.
That is not an assumption I would want to make. There are plenty of accounts out there of sporadic staffing problems at emergency call centers.
It appears the right way to do this is to get an appointment. More details here:
In the UK you need to email BT (who operate the emergency system) and arrange a window to place a series of test calls to both 999 and 112. They will give you a time, a date and a script to exactly follow. If you pre-schedule with them it's supported.
Their email is 999testcalls@bt.com
Not sure if US have a system like it.
Google already special-cases it in the dialler.