We are testing a 4-process WCF IIS application (x3 version) for memory stability (leakage), just pinging it every ~ 1s as a load balancer. It works fine for> 12 hours if nothing works on the server.
However, if we intentionally reduce the total available memory (correct the page, reduce the physical memory, launch other applications) and click on the physical memory usage to 97% and leave it there for 5 minutes or more, often Windows will feel the state and turn off one of processes. Note that it also fails if total memory is pressed up to 97% (using a fixed page file).
However, analyzing the memory size of one of the surviving processes using the RedGate tool shows the following:

Since requests are just persistent ping, there seems to be no practical reason for .NET to keep 269 MB of free memory when the server goes hungry. Approximately 50% of IIS processes are in this state (~ 1.8 GB).
Application compiled against .NET 4.0, gcServer true. The IIS Gate check was set to 0% (minFreeMemoryPercentageToActivateService = "0"), although we probably set it to 2% in production.
Server 2008 R2, 4 GB of physical 4 GB of fixed page, was tested with 4.0 and then 4.5.1 (it doesn't matter).
There is an answer to a similar @atanamir question requiring:
".NET , ".
- - ? ?
: