Thanks for the reply Magneto.
It’s definitely useful in our case - but then it’s for an experiential application so we’ll have control over the hardware and can afford the extra rendering cost.
Still it would be nice to understand why it’s happening and whether it’s something that could be reduced, hacked or potentially fixed in the future.
Hopefully someone else will have an answer to what’s happening…?
I’ve got the same problem - I ended up doubling all the geometry and mirroring it. This wouldn’t work with much animation though, and I ended up with tons of polys.