All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=600508

Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?keys=agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgICgiJWyoAoM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgICgiJXOsQoM


Bot(s) for this bug's original alert(s):

chromium-webrtc-trunk-tot-rel-mac

CL range:
http://test-results.appspot.com/revision_range?start=384283&end=384286

I don't see any CLs related to WebRTC in that range. kjellander@, this doesn't seem to align with when the mac testers were reimaged, but can you take an initial look? If you agree, then I guess the next step is to send to this mflodman@ for additional triage.

Right, the machine was upgraded to 10.10 two days earlier. I have no idea what could be causing this.

I think it looks like it matches the signature for https://bugs.chromium.org/p/chromium/issues/detail?id=600503 quite well, so it's likely that the denoiser was turned off and on again.

Since this is on the Mac FYI bots, the bots detected Per's change earlier than the regular bots (which detected it on the roll a couple hundred chromium revisions later).

However, there's a smaller increase around 383460-383783, which we should investigate. It matches this roll:
matches this roll: 

https://chromium.googlesource.com/external/webrtc/trunk/webrtc.git/+log/df760cd..ce55976

ce55976 Reland of Added a bitexactness test for the gain controller in the audio processing module. by peah · 7 days ago
b9c001f Remove accidentally readded webrtc/base/sslstreamadapterhelper.cc by Henrik Kjellander · 7 days ago
12cfa58 Remove all uses of the HAVE_CONFIG_H define. by Henrik Kjellander · 7 days ago
9d541b1 [rtcp] SenderReport::Parse updated not to use RTCPUtility by danilchap · 7 days ago
893fb1f Move to x509 v3 as required by the WebRTC draft. by torbjorng · 7 days ago
8d360b6 Remove orphaned files. by torbjorng · 7 days ago
91b510a Added a bitexactness test for the beamformer in the audio processing module by peah · 7 days ago
e884401 Android HW decoder: Add support for textures when using EGL 1.0 by magjed · 7 days ago
908eeb9 Change include in metrics.h (change to use systems_wrappers/include/logging.h, base logging breaks chromium.fyi). by asapersson · 7 days ago
7274c0a Fix typo in FakeAdmTest.TestProcess name. by Peter Boström · 7 days ago
63b8d3f Add macros for ability to log samples that are added to histograms (RTC_LOGGED_*). by asapersson · 7 days ago
d27854b Allow passing in strings of length zero to FileWrapper::Write without closing the file. by terelius · 7 days ago
7a29e63 Re-reland of Added a bitexactness test for the echo canceller in the audio processing module. by peah · 7 days ago
019caf7 Fix PeerConnectionInterfaceTest.CloseAndTestStreamsAndStates by Per · 7 days ago
7766e51 Added a bitexactness test for the intelligibility enhancer in the audio processing module by peah · 7 days ago
770073f Build dynamic framework with podspec for Objective-C API. by Jon Hjelle · 11 days ago
08ef89b Revert of Added a bitexactness test for the gain controller in the audio processing module. (patchset #3 id:60001 of https://codereview.webrtc.org/1812433002/ ) by peah · 11 days ago
d79962e Added the JNI interface to get and set RtpParameters and the maximum bitrate limits. by skvlad · 11 days ago
5c481f7 Added a bitexactness test for the gain controller in the audio processing module. by peah · 11 days ago
d4e1bb9 Adding BlockMeanCalculator for AEC. by minyue · 12 days ago
c92ec0a Fixed a potential deadlock problem in the AGC by peah · 12 days ago
b120b5b Avoid clicks when muting/unmuting a voe::Channel. by solenberg · 12 days ago
57abdbf Don't override curve preferences in BoringSSL. by David Benjamin · 12 days ago

The above causes an increase of 1-2% encoder cpu usage. I don't see any obvious culprits in the list though?

This has recovered (and subsequently half-way regressed again), so I think we can declare this obsolete.