Discover the actual/prospect usage matrix to drive testing based on risk:(Sample Matrix)
Testing is purely risk driven activity. I would suggest to find out the production usage base matrix (OS/Browser/devices wise) to find out the actual risk to drive the test efforts accordingly.
It only makes sense to test on wider combinations of configurations if the ...
Depends on the goal of the test.
I like to differentiate test goals, for example between:
Pure Functional, Tests which check if the functionality works in general. This should work on a stable pipeline, simple and fast for quick feedback loops.
Browser/OS/Device compatibility, tests to garantuee it works on multiple different devices.
Now you need to ...
It depends on the context.
If your users are using different browsers and operating systems, or desktop/mobile versions of your app, or different versions of browsers, then it is a good idea to test on multiple build agents.