in this thread

https://boinc.berkeley.edu/dev/forum_thread.php?id=10957

I have reported about my problem to get GPUGRID tasks downloaded while running serveral ATLAS tasks, with the BOINC event log notice:
Long runs [8-12 hours on fastest card] needs 2146.64MB more disk space. You currently have 5482.75 MB available and it needs 7629.39 MB

In the GPUGRID forum, a forum member was mentioning "But I wonder if GPUGrid really needs all that much space? I doubt it for the GPU client; it may be an artifact of the CPU work they were doing at one time. Maybe you could post separately about it, and it could be reduced."

I looked it up closer, and in the BOINC directory, the projects folder for GPUGRID is about 90MB big, the slots folder is about 48MB. So, no idea why when downloading a new task, GPUGRID requests 7.629MB of free space. The assumption about outdated "historical" reasons may be correct.

I wonder whether this stumbling block could be circumvented by some hand-written config.xml or whatever ??? Maybe an interesting challenge for the specialists here?

Unfortunately, I think not. The disk space requirement is set in the workunit meta-data: the current one on this machine is

<workunit>
    <name>e24s5_e17s9p1f222-GERARD_A2AR_OSTRUC_NUL2-0-2-RND4822</name>
    <app_name>acemdlong</app_name>
    <version_num>848</version_num>
    <rsc_fpops_est>5000000000000000.000000</rsc_fpops_est>
    <rsc_fpops_bound>250000000000000000.000000</rsc_fpops_bound>
    <rsc_memory_bound>300000000.000000</rsc_memory_bound>
    <rsc_disk_bound>8000000000.000000</rsc_disk_bound>
    <file_ref>
        <file_name>e24s5_e17s9p1f222-GERARD_A2AR_OSTRUC_NUL2-0-LICENSE</file_name>
        <open_name>LICENSE</open_name>
    </file_ref>
    ... many more files
</workunit>

The value you would want to over-ride is

<rsc_disk_bound> 8,000,000,000

8 (decimal) gigabytes. Per task.

If disk usage was associated with the application, you could re-define it in an app_info.xml: but because it's data, it's correctly assigned to the researcher to configure.