In my DSL testing (with stable base line RTT) BBR slows down just before the connection is saturated (provided its single threaded), so would less likely cause buffer bloat. They have succeeded in making it resistant to packet loss as well as reducing the over buffering problem.
However I assume if you start piling in the threads buffer bloat would still be a problem (but not tested).
So e.g. if I download a file using cubic, we assume line is healthy no background packet loss, throughput graph auto sizes. I would see flatline max speed. (in reality it will be see sawing up and down). Then I download the same file again on BBR with same max scale on graph, the throughput will be slightly below max, there will be a visible but very small unutilised part of the connection. So from a QoS perspective it should be better.
But in unideal conditions for cubic, BBR would outperform it, and then more likely have buffer bloat? As more data sent down the pipe.
I am pretty sure that chrome is also using QUIC now out of the box.
Interesting that litespeed blog page they discovered the same as when I was diagnosing my VDSL issues.