I mentioned this behavior in a thread a long time ago in a galaxy far away. The problem indeed was on my end. I am running Linux Mint. There is a two pronged approach to help alleviate the problem 99% of the time.
First: Make sure the way your connectivity with the internet is stable. It appears that if the connectivity is broken, even for short times, it will cause the WU to display a "signal 11 client is dead" message. The work units are pretty resilient and do recover from these, but there appears to be a hard limit after which BOINC will kill the job.
Second: For those jobs which appear to be running in an endless loop, suspend the job, wait about minut so another job can get a good start and then resume the unit. Once it begins running again when a slot opens up, it will take about a minute to get its bearings once again, and should start running normally.
I used to have these items both occur quite regularly until I gave my range extender a static IP address instead of letting the router give it a DHCP address. Since then I have only had an isolated incident of a running unit spinning its wheels and very little problem with intermittent connectivity glitches.
Hope this helps
Cheers

I changed all changes on BIOS back to "Auto" and "Normal", but got two more WUs with problems!

Paused them both and the scheduler change to a another project, then they began to go crazy: percentage jumped from 83,333% to 1234% and back (and also other numbers)...
Real bad

LAZA74, When you have one of these workunits does it use cpu time in top or ps? Also can you post the stderr log from one of these you have aborted.

Thanks,
armstrdj

I have these very occasionally. I finally caught one on MCM1. It was running on Linux Mint 14 32 bit on a Core2 Duo. Here is the Stderr_txt file.
It is valid. When I saw it was running (accruing time)but not showing any progress and the one core was idle I suspended the WU and then resumed it. It then started running normally. I presume if I had not caught it, it may have run indefinitely.
Result Log

Result Name: MCM1_ 0004153_ 1465_ 2--
<core_client_version>6.10.59</core_client_version>
<![CDATA[
<stderr_txt>
Commandline = ../../projects/www.worldcommunitygrid.org/wcgrid_mcm1_7.32_x86_64-pc-linux-gnu -SettingsFile MCM1_0004153_1465.txt -DatabaseFile dataset-17_72_SDG_v1.txt
Settings File
DateOfDesign = 11/08/2013
Designer = PMCC_OCI
WorkOrderID = 4153_1465
DatasetID = 17_72_SDG_v1
NumberOfGenesInStartingSignature = 19
NumberOfGenesInSignatureMin = 10
NumberOfGenesInSignatureMax = 20
GroupVectorValues = {A}{B}{C}{D}{E}{F}
ExplicitStartingGeneSignatures = A B D F
StartingGeneSignatureAlgorithm = randomFixedLengthSearch
SearchAlgorithmNumberToCreate = 1
SearchAlgorithmSequentialStartPosition = 5
RunPermutationAlgorithm = 1
PermutationGroups = A
PermutationGroupsForReplacement = G
PermutationAlgorithm = replaceFromRandomlyToRandomlyGreedy
PermutationsNumIterations = 54959
OptimizationAlgorithmFrequency = 0 0 1
FBeta = 1.5
SimAnnealIMax = 20000
SimAnnealAlpha = 0.9996
NReps = 10
TrainFrac = 0.7
NFolds = 10
VMethod = LOO
ModelType = SVM
FitnessFn = 0
MinFitness = 0.61
SvmArgs = "-v 0 -c 0.1 -t 1 -d 2 -r 0"
SvmLearnLimit = 500000
RSeed = 400891465


[22:29:00] Initializing
[22:29:03] Running
[22:29:03] EvaluateFitnessOfStartingGeneSignatures 1
[22:29:03]: Computing pass 0
04:22:01 (14403): No heartbeat from client for 30 sec - exiting
04:22:01 (14403): timer handler: client dead, exiting
Commandline = ../../projects/www.worldcommunitygrid.org/wcgrid_mcm1_7.32_x86_64-pc-linux-gnu -SettingsFile MCM1_0004153_1465.txt -DatabaseFile dataset-17_72_SDG_v1.txt
[20:01:15] Initializing
[20:01:18] Running
[20:01:18] EvaluateFitnessOfStartingGeneSignatures 1
[20:01:19]: Computing pass 0
[21:26:14] Exiting PermutateGeneSignature
[21:26:14] Writing final output
[21:26:15] Closing Output Stream
[21:26:15] Cleaning up
Result.out = 2444912.000000
Run complete, CPU time: 26090.620000
21:26:15 (16178): called boinc_finish

</stderr_txt>
]]>

Cheers