Since I stepped from 4.4 to 4.13 with every major release, including 4.10 and the screen problem remained, but the HWE command did solve the issue, it's something outwith the kernel i.e. the sampled command on the HW front that seems to have fixed things. At any rate... humming along in seeming stable performance, no tailing off.

Some googling suggests there were issues with the early 16.04 v.v. I7 6700

I have an observation which is not scientifically rigorous, but appears on the surface to be interesting. I have a 2600K system with 8 gb of memory running Linux which had running both SCC and MIP. The average time gradually rose over the days and appeared to settle about 3.5 to 4 hours per MIP unit. I switched it over to all MIP to see if I had any of the symptoms described by KerSamson. Once it was doing only MIP work the average work unit was taking 5.5 to 6.5 hours. I thought maybe it was the result of just bigger workunits. I also have a system running all MIP units, a dual Xeon E5410 with 8gb of memory also running Linux. This machine should be slower than the 2600K, but even when the 2600K was running the longer units, the Xeon continued to run units in the 4 to 4.5 hour range with only an occasional unit being longer.
I don't know what exactly is happening, but suspect the Xeon is a bit better at running the MIP units than the 2600K. There must be some piece of the architecture in the 2600k which is being overloaded versus the older Xeon. The Xeon may have some part of its architecture which is less prone to being maxed out than the 2600K. It may have something to do with the 2600K being a consumer chip and the Xeon being a server chip.
When both machines were running SCC exclusively the 2600K was notably faster by about 15 to 20 minutes per work unit. This is the only time I have noticed this kind of difference in the performance of the two machines. I have since switched the 2600K back to SCC.
Just an observation.
Cheers

Yes, I have hyperthreading turned on in the 2600K. You could be right on either the hyperthreading issue or the L2 cache difference or may a combination of both items. MIP does something different than SCC and so far MIP is the only project where the Xeon is faster than the 2600K. Thanks for the input.
Cheers

can anyone say something about the checkpoints? 1 checkpoint after <10minutes and the next after another hour?! I crunch with a private computer, so I have to suspend it or are there other tricks?

I think the checkpoints are hard coded into the application, so there is probably nothing you can do to change when they occur.(Someone correct me if I am wrong.) You can check all the projects and run just ones that have checkpoints which occur at regular short intervals.
Cheers