Just had a quick look at crash/ - two observations.

First - all the "out of memory" buckets for browser have the right exception code, so the code looks like it's working.

Second - adding up all "Out of memory" from crash/ for browser process, stable channel, gives approx 4.5% of crashes are OOM, but looking at the UMA metric - 536870904 is only 0.01%.

So, there is something definitely wrong with that UMA metric. It's worth some investigation (i.e. simulate an OOM in browser - see what happens). I'm happy to take this as a bug assigned to me (since I already wrote chrome://memory-exhaust for renderers).