Bisecting Chrome between 61.0.3142.0 and 3146.0 using Dean's script showed that the regression is introduced by this CL:

commit 56679d20bd6284ffe2a64224d33232cb1df4f0ba
Author: Xiyuan Xia <xiyuan@chromium.org>
Date:   Thu Jun 29 21:44:41 2017 +0000

    cros: Fix user policy autotest failures
    
    Skip policy token fetching when "--disable-gaia-services" switch
    is present. Otherwise, chrome would not talk to fake DM server
    to load testing policy and fail test expectations.
    
    BUG=b/62523541
    
    Change-Id: I0859def40f343b7ecdd05b9b147b35bfa3705ead
    Reviewed-on: https://chromium-review.googlesource.com/555654
    Commit-Queue: Xiyuan Xia <xiyuan@chromium.org>
    Reviewed-by: Maksim Ivanov <emaxx@chromium.org>
    Cr-Commit-Position: refs/heads/master@{#483499}

Instead of a real browser startup time regression, the longer test time seems due to retry timeouts at the gaia login stage (like the one Achuith fixed earlier) as the test is stuck at the login phase for 180 seconds with this command in chroot:

test_that <IP_OF_DUT> --args local=True telemetry_Benchmarks.startup.cold.blank_page

Xiyuan, could you take a look?

I'll take a look, since Xiyuan is OOO.

We have the same regression findings (but on other board): crbug.com/739107