IMHO a simple multimeter display really doesn't need the price overhead of WinCE.
But if that is all the manager who's decision it was knows then it was probably the better choice.
Who thinks this is a "simple" multimeter?
It has USB interface, file systems, Ethernet, a huge graphical user interface display that does all sorts of stuff etc.
For a product like this, the need to use an OS is marginal - there are plenty of standalone graphics libraries, the UI is just buttons. USB functionality is very limited, and filesystem is not a huge deal either.
If you were just designing this product in isolation, then using any 3rd party OS could probably be avoided with similar development effort to learning and adapting one.
However for a company like Agilent, they need something that's useable on a wide range of architectures, including products that really need a complex OS.
If the OS was designed in a way that allowed it too be easily stripped down to only what was needed for a particular application it wouldn't be an issue - what does the DMM OS actually _need_ to do at startup? Almost nothing - just some selftests.
I'm convinced that the bulk of boot time in OSs like this is nothing but lazy programming - inappropriate timeouts, trying to talk to hardware that isn't there, decompressing code that's never needed, not threading things like network initialisation properly etc.
It would certainly be possible to write an OS for this sort of application with negligible boot time.
The problem is how many people consider startup time a sufficient issue to be a major selling point? Unfortunately very few, so there is no incentive to invest in improving things.
This is a problem with the majority of embedded systems - many modern LCD TVs take longer to turn on than the old CRT ones.
BTW I hear on good authority that the priority for this product was getting something out the door that worked properly, and there is a new firmware version in the works which is likely to approximately halve the startup time.
However annoying boot time is, Software development time is always finite, and given the choice between slow boot and flakiness in operation, the former is the least-worst option.