I said that because I had WUs all day long. Most days I get so few that I don't see the ones I get.

what do you mean "keep your card running" in order to get them? the looping command just keeps making BOINC ask for more work using a tool that is shipped with BOINC. it doesn't make the GPU do any work just asking for work. that's all in the BOINC scheduler. you can absolutely crunch other projects while letting the looping command run in the background. if/when you get a successful request with some work, you'll crunch them, send them back, then continue to crunch your backup broject while waiting for more work. same as any other.

I don't agree that you can think of it as "taking" tasks away from anyone. no user is more deserving than anyone else from the project's perspective. they have work to do and a pool of volunteers. if a willing participant gets the work and intends to complete it, no issue.

did you look at GPUGRID? they are another GPU project in the biomedical space. although they don't have consistent work right now. they do have some beta tasks that pop up in small batches. but make sure you have the system resources to handle it. you'll need 16GB of system ram PER GPU task.

BOINC gives you the ability to set exclusive applications, which will pause BOINC processing when it detects that another defined process is running. I have to assume that folding gives you the same capability, no? a 2 second google search revealed this, though third party. https://foldingforum.org/viewtopic.php?nomobile=1&f=14&t=37076.

set the opng application (not boinc) to be the trigger, and whenever folding detects that application running, it'll pause folding. and resume when the application exits. this way you can run the looping command via boinc in the background, process the few opng tasks when they show up, then resume folding in the downtime.

Look at them? I have been doing them for years.
See one of the stickies.

They haven't had work for months, except a few test batches.
I don't think you have considered this carefully.

Every time I look at GPUGRID’s server stats page there are some tasks in progress. Work has to be available for it to be in progress by others.

They have work. Not a lot. But they have it. Claiming they haven’t had work for months is objectively false. You’re either not requesting work regularly, or you’re not interested in their machine learning based beta work. It’s fine if you’re not interested in that work, but that doesn’t mean it’s not available.

But this goes back to the original statement about opng work. If you want it, you have to ask for it regularly. And the most effective way to do that is with a looping command or script, due to the limitations in BOINC’s scheduler logic.