Got 24 total (22 binaries + 2 resends), all proceeded w/o problems.

OK, have had a look around the farm.
* All machines are Win7-x64 Pro EXCEPT for 1 x 2600K with XP-x64, BOINC 7.0.64 x64, service install, LAIM ON.
* I am logged in continuously on all machines.
* All machines are overclocked but have run for many months without any errors related to that.
* [Edit] The 2600K with XP-x64 has a flaky CPU that gives occasional nasty errors if fed CEP2 WUs - last one required a Windows repair-re-install. This machine does not now run CEP2, but all others are always running CEP2 WUs. [/Edit]
* All machines have mechanical hard drives that are in IDE mode, not AHCI.
* All v7.22 WUs have returned Error, with "too many exit(0)s", except for [Edit] 1 WU that ran OK and is PV on the 2600K (XP-x64 and no CEP2 running!)[/Edit], plus the 2 WUs I still have suspended on the '970 as per my post above.
The 2 v7.22 WUs on the 970 were also looping with exit(0)s and restarts when I suspended them.
I will hold them suspended a bit longer in case someone has a suggestion of something to try with them.
Otherwise I will resume them, and expect they will error out when they reach the limit on the number of exit(0)s allowed.

I don't think these exit(0)s are a startup problem like the ones that can happen when too many troublesome WUs like CEP2 ones are started together.
These exit(0)s happen when only 1 is started, I have not noticed the system freeze, and there are no "no heartbeat" messages.

Now, looking around my Results Status pages has shown a few things that could be of use to the techs:

1. The result logs of my Error WUs have the following group of lines that repeats every 10 sec until the limit on the number of exit(0)s is reached:

> Commandline = projects/www.worldcommunitygrid.org/wcgrid_beta17_7.22_windows_x86_64 -SettingsFile BETA_9999988_0347b.txt -DatabaseFile dataset-GDS2771-v1.txt
> Initializing
> wcg_learn_limit = 100000
> Running
> INFO: WcgLearnLimit(100000) reached. 0.0015534841631961 0.0015534841631961
> 11:49:07 (2924): BOINC client no longer exists - exiting
> 11:49:07 (2924): timer handler: client dead, exiting

Relevant info:
* It's a 64-bit program
* WcgLearnLimit is reached very quickly
* Claims that the BOINC clients are dead are completely false. They are all still very much alive and kicking!

2. In many cases, wingmen have completed the WU successfully and their copies are PV or have validated.
But there are variations:
Example 1: BETA_BETA_9999988_0347b (Mine has too many exit(0)s, the 2 other wingmen have validated)
Wingmen _0's log file (_2's is similar):
> 0000) reached. 0.0022961953632290 0.0022961953632290
> INFO: WcgLearnLimit(100000) reached. 0.0015725667553852 0.0015725667553852
> INFO: WcgLearnLimit(100000) reached. 0.0204744401353310 0.0204744401353310
... then many similar lines down to ...
> INFO: WcgLearnLimit(100000) reached. 0.0025786576999280 0.0025786576999280
> Result.out = 41760.000000
> Run complete, CPU time: 5070.079300
> 19:15:39 (10084): called boinc_finish

No records of their commandlines. Maybe the first parts of their logfiles have been truncated for display on the website.

Example 2: BETA_BETA_9999987_0122b (Mine has too many exit(0)s, the 2 other wingmen have validated)
Other wingmen's log files are similar to:
> <core_client_version>7.0.28</core_client_version> ((the other wingman has 6.10.58))
> <![CDATA[
> <stderr_txt>
> Commandline = projects/www.worldcommunitygrid.org/wcgrid_beta17_7.22_windows_x86_64 -SettingsFile BETA_9999987_0122b.txt -DatabaseFile dataset-GDS2771-v1.txt
> Initializing
> wcg_learn_limit = 250000
> Running
> [22:33:41]: Computing pass 0
> [23:28:52]: Computing pass 1
> Result.out = 10036995.000000
> Run complete, CPU time: 5668.545937
> 00:08:39 (532208): called boinc_finish

So all 3 of us were running the 64-bit Windows beta program.
Earlier, I found some apparently successful runs where the wingmen ran 32-bit programs and thought it might be a 64-bit-only problem. Not so.
No info on wingmen's service- vs user-installs.

[Edit] NEW INFORMATION - See "[Edit]"s above.
The WU that ran OK on the flaky 2600K was BETA_BETA_9999986_0879b.
It was a rather short WU, the last lines of its log file being
> [11:27:35]: Computing pass 472
> Result.out = 9062541.000000
> Run complete, CPU time: 819.265625

Maybe the problem relates to running CEP2 WUs at the same time as these betas ... Comment, anybody??

[Update]: No CEP2 WUs are now running on the 970 (1 is suspended), so I resumed 1 of the 2 suspended v7.22 betas. It still exits and restarts, so I don't think CEP2 is the problem.
I think it's Windows 7, while XP-x64 is OK.
[/Edit]

If not mistaken[I have a historical misremembering on this item **], the [BOINC] limit is 100 restarts before the plug is pulled on a task, ending in the very message of "too many exits...".

** So I checked, and yes it is 100 times, to quote from another project "I think climateprediction fell victim to the "100x Task exited with zero status but no 'finished' file" bug"

@nanoprobe, if you're reading, you install IIRC as user to enable the GPU part of the client, which else would not work on Windows.

edit: errata

Although I still have some ongoing WUs plus some in the PV state, all seem to run as they should. I didn't get a chance to test them a lot though, just some suspending a WU or put the laptop to sleep (LAIM on) and there were no apparent problems there. I did spot a few "if this happens repeatedly..." messages in one of the logs, but those WUs eventually completed successfully and were uploaded (maybe even validated already). Either way, looks like a much better run than the last one.

Hope this helps. Cheers!

Version 7.22 seem to be running OK.
BETA_ BETA_ 9999982_ 0617b_ 1-- restarted correctly after an overnight shutdown. The stderr on the results status page is incomplete as the last recorded 'computer pass' message is at 07:47 but the work unit completed at 10:09.
I also had the impression that on two other work units that the elapsed and CPU times decreased when the progress moved from the static 0.5%. On the fourth work unit I missed the move from 0.5% after about 40 minutes.