I don't know the details of your GPU, but I was facing this on one of my systems a few months back. It turned out that the power supply in the computer wasn't up to the task, and when the GPU loaded up with the OPNG WUs, the computer wouldn't fully hang, but the screen would go black, and the GPU would become nonfunctional until a restart. New(er) power supply sorted it out, and it runs properly now.

I *think* I've also had this happen when I try to remote into the systems (with RDP, to either Windows or Linux hosts). It seems to do something with the GPU rendering sent over the network that messes up local access to the GPU :/ I don't have enough firm data to claim this concretely though.

My old but still working GTX660M crunches these 80+ "jobs" WU's, in a little bit over 6 hours.
Example: https://www.worldcommunitygrid.org/contribution/workunit/280054486

Still worth running it since the BOINC credit /WCG points those WU's gives/hour, is far higher than an extra CPU core can produce in the same number of hours.

My also old GTX980 Strix, will crunch two 80+ "jobs" OPNG at the same time, in around 15 minutes.
But until there are plenty OPNG tasks available, I'll let that computer rest.

So of course, as soon as I say all are failing one runs through to completion! It took 11 hours but it got there in the end.

I’ve reloaded the Nvidia drivers on the other machine and confirmed that OpenCL is now running so it’s now just a case of waiting until both machines download more OPNG to see what happens.

Is the “time limit exceeded” based solely on processing time or has the WU’s “Deadline” time been passed?

If it is the latter, you might check your queue values to make sure that setting is not too high causing the WU’s to sit idle until it is their turn to be processed. Meanwhile, the clock tics on and expires.

No, purely based on the runtime.