Any progress on this. I've been thinking about it too. Couple of ideas:
Too many indexes needing to update when an insert occurs?
Are there any triggers running upon insert?
Unlikely but there isn't a disk write bottleneck? Might be worth running some benchmarks from the VM shell.