dbus-broker vs dbus-daemon for the accessibility bus

[thomasw@fastmail.cn]

For those not in the know, dbus-broker is a reimplementation of the dbus-daemon.
I am very interested in this subject when it comes to the accessibility bus. I have done two types of tests on debian. This requires building the broker from source and rebuilding at-spi2-core. Used cpu was an 8550u.
Test 1: Load large pages in Chrome with its accessibility enabled with Orca on. I find that the accessibility bus often uses up to 50 percent of a core with the daemon and 30 percent with the broker. You need to use Chrome 77 or older because a change was recently made to Chrome master which makes it send far less across the bus when loading a page. World War ii page on wikipedia works nicely.
Test 2: Put the daemon and broker processes in a cgroup and limit their cpu. In my unscientific testing, I find that the broker seems to perform a bit better with less cpu. As you choke off the cpu more and more, the gap widens.
Just curious if anyone has done similar testing on debian or even other distros. I have a core i7 8550u so I don't notice a huge difference but I suspect users with less cpu might. I have an older core 2 duo machine. It would be very interesting to get another of those and do side by side testing.