Steps to reproduce the problem:
In an Android workspace:

1. Build the content_browsertests_apk target
2. Connect an Android N+ device
3. Run the memory instrumentation browsertests:
out/Debug/bin/run_content_browsertests --num_retries=0 -f *MemoryInstrumentationTest*

What is the expected behavior?
The one memory instrumentation test (PrivateFootprintComputatio) runs and passes.

What went wrong?
Roughly 9 times out of 10, the test will fail due to the memory footprint not changing after freeing the memory allocated for the test.
e.g.

C   80.863s Main  ********************************************************************************
C   80.863s Main  Detailed Logs
C   80.863s Main  ********************************************************************************
C   80.863s Main  [UNKNOWN] MemoryInstrumentationTest.PrivateFootprintComputatio:
C   80.863s Main  [ RUN      ] MemoryInstrumentationTest.PrivateFootprintComputatio
C   80.863s Main  [WARNING:process_metrics_linux.cc(708)] Failed to open /proc/vmstat
C   80.863s Main  [ERROR:devtools_http_handler.cc(292)] Cannot start http server for devtools. Stop devtools.
C   80.863s Main  [WARNING:child_process_launcher_helper_posix.cc(118)] Ignoring invalid file assets/snapshot_blob_64.bin
C   80.863s Main  [ERROR:shell_android.cc(78)] Not implemented reached in void content::Shell::PlatformSetTitle(const base::string16 &): Error
C   80.863s Main  ../../content/browser/tracing/memory_instrumentation_browsertest.cc:105: Failure
C   80.863s Main  Value of: after_kb - before_kb
C   80.863s Main  Expected: (is >= -6656) and (is <= 6656)
C   80.863s Main    Actual: 66824
C   80.863s Main  ../../content/browser/tracing/memory_instrumentation_browsertest.cc:109: Failure
C   80.863s Main  Value of: during_kb - after_kb
C   80.863s Main  Expected: (is >= 63560) and (is <= 69560)
C   80.863s Main    Actual: 0
C   80.864s Main  [WARNING:discardable_shared_memory_manager.cc(436)] Some MojoDiscardableSharedMemoryManagerImpls are still alive. They will be leaked.
C   80.864s Main  [  FAILED  ] MemoryInstrumentationTest.PrivateFootprintComputatio, where TypeParam =  and GetParam() =  (1295 ms)
C   80.864s Main  [----------] 1 test from MemoryInstrumentationTest (1295 ms total)
C   80.864s Main  
C   80.864s Main  [----------] Global test environment tear-down
C   80.864s Main  [==========] 1 test from 1 test case ran. (1295 ms total)
C   80.864s Main  [  PASSED  ] 0 tests.
C   80.864s Main  [  FAILED  ] 1 test, listed below:
C   80.864s Main  [  FAILED  ] MemoryInstrumentationTest.PrivateFootprintComputatio, where TypeParam =  and GetParam() = 
C   80.864s Main  
C   80.864s Main   1 FAILED TEST
C   80.864s Main  <<ScopedMainEntryLogger
C   80.864s Main  ********************************************************************************

It looks like in Android N and beyond, the memory footprint isn't immediately reduced, but the test is expecting it to be.
If 30 second sleep is added after buffer.reset(), the test will begin to pass again (but obviously this isn't a reasonable solution).

When run on Android 5.1, the test passes consistently.
When run on Android 7.1.2 or Android 8.1.0, the test fails nearly every time.

Did this work before? N/A 

Chrome version: Master branch  Channel: stable
OS Version: 8.1.0
Flash Version: 

These tests are not currently being run on Android N in BuildBot, but they are being run on L and M:

* Lollipop: https://ci.chromium.org/buildbot/chromium.android/Lollipop%20Phone%20Tester/20029 (content_browsertests all passing)
* Marshmallow: https://ci.chromium.org/buildbot/chromium.android/Marshmallow%2064%20bit%20Tester/20874 (content_browsertests all passing)
* Nougat: https://ci.chromium.org/buildbot/chromium.android/Nougat%20Phone%20Tester/4861 (content_browsertests not present)