UserAgent: Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/52.0.2743.116 Safari/537.36

Steps to reproduce the problem:
1. Open a tab on Chrome Win64 
2. Open Developer Tools
3. Execute in the console:
    mem = []; for (var i = 0; i < 1024 * 4; i++) {mem.push(new ArrayBuffer(1024*1024))}

This will attempt to allocate 4 gigs worth of ArrayBuffer objects.

What is the expected behavior?
When a script attempts to allocate memory beyond the memory limit (somewhere just shy of 4 gigs, per https://cs.chromium.org/chromium/src/content/common/sandbox_win.cc?q=SetJobMemoryLimit&sq=package:chromium&dr=C&l=578), the allocation should fail, and an exception should be thrown.

What went wrong?
When a script running in a sandboxed process allocates memory beyond the limit, the tab crashes.

Even with logging enabled and --v3, nothing is printed to the debug log when this occurs. Running with --no-sandbox makes it so the issue cannot be observed, as tabs run with no sandbox and thus no memory limit.

Did this work before? No 

Chrome version: 52.0.2743.116  Channel: stable
OS Version: 10.0
Flash Version: Shockwave Flash 22.0 r0

From the Windows API docs: "If a process associated with a job attempts to increase its working set size or process priority from the limit established by the job, the function calls succeed but are silently ignored."

They're not clear about the outcome, but I believe this says that allocations that would increase the process working set beyond the limit will silently fail. 

It's as though they know this is horrifying, so they provide a mechanism to set a memory threshold for the job - when the threshold is reached, a notification will be provided to the offending process.

Job limit docs: https://msdn.microsoft.com/en-us/library/windows/desktop/ms684161(v=vs.85).aspx#job_limits_and_notifications
Job notification config object: https://msdn.microsoft.com/en-us/library/windows/desktop/hh448386%28v=vs.85%29.aspx?f=255&MSPPError=-2147217396

Scope of work:
The sandbox code needs to be modified to set a notification limit on the job object - the notification limit should be some reasonable amount less than the hard limit, maybe 64 megs or so less. When the notification arrives, any operation originating from script code that would allocate additional memory should fail throwing an exception.

Ultimately, it would be nice to be able to set a higher limit via a flag or something. Limiting a process to 4 gigs is a decent default, but it's an arbitrary baked in limit. It seems like bad design, especially since it seems the same limit doesn't exist on other platforms.