Under:
Windows 8 ver 6.3 (build 9600)
Boinc Manager 7.16.11 (x64)
(don't know how to determine Rosetta version)

Several tasks for: pre_helical_bundles_round1_attempt1
are "Waiting for Memory" despite this family of tasks usually using 450-500 MB RAM & there being 6 GB RAM available.
Even if I stop running every other task on my i5-4430 PC & switch CPU usage to 25% (ie. just use 1 of the 4 cores) even 1 of this small no. of tasks refuses to run: "Waiting for Memory" - a single task needing 6 GB of RAM... seems unlikely: 10x more than most of these tasks in: pre_helical_bundles_round1_attempt1

If whoever submitted them would like to check these over-ambitious children, here are a few names:
pre_helical_bundles_round1_attempt1_SAVE_ALL_OUT_IGNORE_THE_REST_3ul8od2k_1389923_1
pre_helical_bundles_round1_attempt1_SAVE_ALL_OUT_IGNORE_THE_REST_6is1nt3i_1389933_1
pre_helical_bundles_round1_attempt1_SAVE_ALL_OUT_IGNORE_THE_REST_6kb0wa5s_1389916_1
pre_helical_bundles_round1_attempt1_SAVE_ALL_OUT_IGNORE_THE_REST_6hr0et4q_1389968_1

I also have low-power PCs running BOINC under Windows 10 which are great at it as they are very power-efficient (10 W) & they have recently taken to saying Rosetta tasks are not being downloaded because there is not enough RAM. It is true that they have 4 GB RAM (2 GB of which runs Windows 10) but the 2 GB RAM is unlikely to be a restriction since Windows 10 is very good at "paging" RAM onto "disk" as work goes along, & since the "disk" is SSD is not wildly slower than the RAM anyway.

My suspicion is that someone has tried to make the task-management algorithms more clever & that some of the tasks submitted either have halucinogenically ambitious "required resources" (particularly for home users) or have simply been mistaken in their parameters.

Obviously, the down-side to having messed-up task-management is that one starts looking at the many other other online science projects that may be more productive...

(I wish this text box was outlined for the user to type in rather than the user having to click randomly on screen where they think it might be - you might like the former English newspaper soccer-photo competition: "spot the ball": where entrants cut out the soccer game photo & mark "X"s where they think the (erased) ball should be, sending it in by post - however, important to read the rubric that says "the winner is the entry nearest where the expert jury believes the ball to be" ie. not where the ball actually was when the photo was taken...)

On the subject of "what resources are being requested", further to my original post, the message that is generated on my little 4 GB RAM low-power PCs is wrong since it complains that "total RAM required is 3957 MB (or something like that) & 6535 RAM is required" - this is daft of course because 3957 MB is the entire RAM of the PC running Windows, ignoring the fact that Windows occupies 2GB of it.

However, rather like code-breaking, this erroneous information may tell us something useful: what RAM the pre_helical_bundles_round1_attempt1 tasks are requesting.
We could assume the task-manager is (erroneously) checking at download that the RAM available is the entire RAM (8 GB) which gets an ok for download for the task, but we know that when it tries to run the actual available RAM is 6 GB which is "not enough" so, donning the deer-stalker we should conclude 6 GB > RAM requested > 8 GB, so "finger in the air guess": 7 GB. For a single task.

Really?

Ambitious would be an entirely fair description.

Under:
Windows 8 ver 6.3 (build 9600)
Boinc Manager 7.16.11 (x64)
(don't know how to determine Rosetta version)

Several tasks for: pre_helical_bundles_round1_attempt1
are "Waiting for Memory" despite this family of tasks usually using 450-500 MB RAM & there being 6 GB RAM available.
Even if I stop running every other task on my i5-4430 PC & switch CPU usage to 25% (ie. just use 1 of the 4 cores) even 1 of this small no. of tasks refuses to run: "Waiting for Memory" - a single task needing 6 GB of RAM... seems unlikely: 10x more than most of these tasks in: pre_helical_bundles_round1_attempt1

If whoever submitted them would like to check these over-ambitious children, here are a few names:
pre_helical_bundles_round1_attempt1_SAVE_ALL_OUT_IGNORE_THE_REST_3ul8od2k_1389923_1
pre_helical_bundles_round1_attempt1_SAVE_ALL_OUT_IGNORE_THE_REST_6is1nt3i_1389933_1
pre_helical_bundles_round1_attempt1_SAVE_ALL_OUT_IGNORE_THE_REST_6kb0wa5s_1389916_1
pre_helical_bundles_round1_attempt1_SAVE_ALL_OUT_IGNORE_THE_REST_6hr0et4q_1389968_1

I also have low-power PCs running BOINC under Windows 10 which are great at it as they are very power-efficient (10 W) & they have recently taken to saying Rosetta tasks are not being downloaded because there is not enough RAM. It is true that they have 4 GB RAM (2 GB of which runs Windows 10) but the 2 GB RAM is unlikely to be a restriction since Windows 10 is very good at "paging" RAM onto "disk" as work goes along, & since the "disk" is SSD is not wildly slower than the RAM anyway.

My suspicion is that someone has tried to make the task-management algorithms more clever & that some of the tasks submitted either have halucinogenically ambitious "required resources" (particularly for home users) or have simply been mistaken in their parameters.

Obviously, the down-side to having messed-up task-management is that one starts looking at the many other online science projects that may be more productive...

In short, you're right that in the last month tasks have been changed from being configured to require around 1.7Gb per task before coming down (even though they often use far less while running) to nearer 6.6Gb before downloading.
And yes, you're right that this makes it very difficult for host machines with comparatively small resources to run anything at all.
We've observed a loss of around 40% of hosts being able to grab anything since the change.
You're by far not the only person to complain (an understatement).

After contacting project admins, we've found out that this isn't a matter of researchers playing fast and loose with RAM (and Disk space) they require but a product of the kind of work they're now needing to run.
This isn't a project that tailors the work they do to the ability of their volunteers to run their tasks. The work they have is the work they have and if people don't have the resources to run them then their hosts are better employed to work with other projects whose work they can run, much as you've guessed, and however disappointing that is for everyone.

You have a couple of choices first though with your constrained RAM hosts:
You can maximise the RAM you make available to BOINC in your Options/Computing preferences/ Disk and Memory tab - if you set it to 99% it seems this only gets looked at at the beginning
You can restrict the number of tasks that run at the same time in your Options/Computing preferences/ Computing tab - eg Use at most 50% of CPUs (meaning cores)

If none of those options work, an alternate project may be your only option.

Project admins have looked again at the RAM that's required and have been able to reduce the 6.6Gb demand to nearer 3.8Gb but, first, it seems we'll have to wade through 18million queued tasks with the higher setting before seeing tasks with the slightly lower setting, though 2nd, this still may be too much for your hosts anyway and it may take 7 weeks for that to be seen as well, such is the rate that the queue is being run down.

Sorry there isn't better news to give you.

---