Here's better test output from running it locally:

DISPLAY=:20 third_party/WebKit/Tools/Scripts/run-webkit-tests -t Default -v --additional-drt-flag=--site-per-process --no-retry-failures --additional-drt-flag=--no-sandbox --iterations=1 svg/dom/getScreenCTM-ancestor-transform.html

This is a testharness.js-based test.
FAIL getScreenCTM: Outermost SVG element ancestor position/transform Cannot read property 'querySelector' of null
FAIL getScreenCTM: Outermost SVG element ancestor position/transform 1 Cannot read property 'querySelector' of null
FAIL getScreenCTM: Outermost SVG element ancestor position/transform 2 Cannot read property 'querySelector' of null
FAIL getScreenCTM: Outermost SVG element ancestor position/transform 3 Cannot read property 'querySelector' of null
FAIL getScreenCTM: Outermost SVG element ancestor position/transform 4 Cannot read property 'querySelector' of null
FAIL getScreenCTM: Outermost SVG element ancestor position/transform 5 Cannot read property 'querySelector' of null
FAIL getScreenCTM: Outermost SVG element ancestor position/transform 6 Cannot read property 'querySelector' of null
Harness: the test ran to completion.

The test in question uses blob URLs in <iframe>s.

Huh.  We shouldn't be creating out-of-process iframes for those, so I wonder what's different.  And the test also runs successfully if I load it via a file:// URL into Chrome when running with --site-per-process.

I'm puzzled why iframe.contentDocument would be null (from the querySelector error in comment 1).

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/07f6ac013d83489dae02a319e0962e666d46ed8e

commit 07f6ac013d83489dae02a319e0962e666d46ed8e
Author: creis <creis@chromium.org>
Date: Tue Feb 28 19:41:02 2017

Test expectation for getScreenCTM-ancestor-transform.html

BUG= 697111 
TEST=Site Isolation Win FYI bot goes green.
NOTRY=true

Review-Url: https://codereview.chromium.org/2720093003
Cr-Commit-Position: refs/heads/master@{#453670}

[modify] https://crrev.com/07f6ac013d83489dae02a319e0962e666d46ed8e/third_party/WebKit/LayoutTests/FlagExpectations/site-per-process

AFAIK, the WPT html5lib tests use a similar "technique", so might be worth checking if the failure mode is the same there (I saw them listed in the flag expectations for site-per-process.)

It looks like this boils down url::Origin's opinion on what is "unique" (has  opaque origin), through its use in RenderFrameHostManager::CanSubframeSwapProcess.
When using the test runner (--run-layout-test) we'll get blobs like "blob:file:///<uuid>", with an origin of {"file", "", 0}, while outside the runner it'll be "blob:null/<uuid>" (for something served from file:). url::Origin consider the former to not have an opaque origin while the latter would. While file: is a bit of an odd-ball here, wouldn't this also have a effect on other blob: URLs with "valid" URLs (paths)?
Anyway, I don't see my test being at fault here, so I'll reassign this for now.

Thanks!  Alex, sounds like your theory may have been close (i.e., layout tests acting differently than outside the runner).  Since you're looking at CanSubframeSwapProcess for the transfer bug, can you take a look at this as well?

Unblocking  issue 678167 .

I think that this is actually a product bug - SiteInstance::GetSiteForURL tries to take blob (and filesystem) URIs into account, but this doesn't quite work for blob URIs associated with a file: origin.

WIP CL @ https://crrev.com/c/953129

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/217fd27e37476bf77cf6b789017d1285e82664f3

commit 217fd27e37476bf77cf6b789017d1285e82664f3
Author: Lukasz Anforowicz <lukasza@chromium.org>
Date: Wed Mar 07 21:41:43 2018

Correctly handle blob:file:///... URIs in SiteInstance::GetSiteForURL.

file URIs should map to "file:///" site.  The same site needs to also
be used for blob:file:///... URIs - this is what is fixed by this CL.

Bug:  697111 
Change-Id: Iff9e9a1552eb747ecd576aefdc5085009588690b
Reviewed-on: https://chromium-review.googlesource.com/953129
Reviewed-by: Nick Carter <nick@chromium.org>
Commit-Queue: Łukasz Anforowicz <lukasza@chromium.org>
Cr-Commit-Position: refs/heads/master@{#541589}
[modify] https://crrev.com/217fd27e37476bf77cf6b789017d1285e82664f3/content/browser/site_instance_impl.cc
[modify] https://crrev.com/217fd27e37476bf77cf6b789017d1285e82664f3/content/browser/site_instance_impl_unittest.cc
[modify] https://crrev.com/217fd27e37476bf77cf6b789017d1285e82664f3/third_party/WebKit/LayoutTests/FlagExpectations/site-per-process