Good point... perhaps there will be a bit of UI to the tests.
E.g. first a simple press a key.. if it isn't pressed within 10 seconds the test fails.
Then for each controller.. press and release all buttons (one at a time). The test can pass once all buttons have been observed in both the pressed and unpressed states. Fail with a timeout, or maybe 'Esc' detected via AWT.
Then try each analog axis, confirm more than just 2 values read back from them.
Not perfect, actually not much better than ControllerReadTest.. hmmm. and actually since my Powerbook reports 2 mouse buttons but has only one mouse button it would always fail

argh...
I'll try to get the ControllerReadTest going on my Linux box tonight. I will only be able to test mouse and keyboard.