Is there any possibility to set/change in the BOINC manager (or somewhere else) the interval in which BOINC is automatically requesting new tasks from either a specific project or all projects?

Not directly. The basic principle is "it asks when (it thinks) it needs new tasks". There are some settings which influence that, but before we dig them all out, could you expand a bit more on what you're hoping to do - what change do you want to see?

okay, I'll try to explain:

beside other projects, I am running GPUGRID, where, as you may know, since beginning 2016 thay have only one scientist left who feeds the WU queue.
This results in a chronic shortage of new tasks.

I have made the experience that when trying to get a task "manually", i.e. by frequently (i.e. every two minutes or so) pushing the "update" button for GPUGRID in the projects tab of the BOINC manager, the chances are somewhat higher.
However, it's clear that I cannot sit there all the time and push the "update" button in 2-minutes' intervals for hours :-)

So what I am trying to find out is whether there is a possibility to set a shorter interval for "requesting new tasks" as it is when BOINC does it by itself (I guess, on the average every 10 or 15 minutes or so, from what I think I saw).
I have realized that by NOT trying to get a task manually, it could take up to 10 or 12 hours to finally receive a new WU :-(

However, it's clear that I cannot sit there all the time and push the "update" button in 2-minutes' intervals for hours :-)

Ah yes, been there. Still doing that. Haven't got the T-shirt yet (though I do have a task on every eligible computer at the moment).

I have realized that by NOT trying to get a task manually, it could take up to 10 or 12 hours to finally receive a new WU :-(

No, there isn't a set interval. The client will request, the first time, as soon as it feels the need. But if it requests work, and receives none, the interval between requests will increase: BOINC will back off, longer and longer, between requests. The idea is to avoid projects with large numbers of volunteers and little work creating their own DDOS attack on the server, meaning that nobody can get through and the project gets no work done - a lose-lose situation.

So, what can you do? Obviously BOINC itself would say 'let another project run in the gap between GPUGrid tasks' - after all, BOINC was created to make scientific use of idle resources.

If you are specifically interested in the *cough* scientific research done by GPUGrid, you can automate the update button using the boinccmd tool:

boinccmd --project URL operation

--project URL operation
Do operation on a project, identified by its master URL. Operations:
...
update: contact scheduling server;

and using your operating system's task scheduler to repeat when necessary. I'd suggest every five or ten minutes, rather than two - don't overtax the server.

Finally, BOINC is open source - I found the build instructions for Windows using the free 'Community Edition' of Microsoft Visual Studio were remarkably easy to follow and successful. I run a self-compiled client with one of the more irritating features disabled to help me debug some project problems: it would probably be fairly easy to find the increasing backoff code and disable that too, replacing it with a fixed value. That would mean that you wouldn't keep pestering the server after you'd snagged a task, as the boinccmd suggestion would.