Model instancing was implemented some months ago, but never made it into a release. We found that the speed gain in normal levels was too small for being a useful feature.