It sounds as though the same issues that make it perform suboptimally on VM hypervisors would make it also perform suboptimally for OLTP databases -- in both, the I/O patterns generally involve high numbers of small writes.