There's no reason "virtualized" (user-submitted scripts?) can't be almost as fast as native contracts. This tradeoff concept exists because ethereum made some design decisions which are good for thin clients but bad for performance (there are sequential crypto operations everywhere, in and out of transaction logic. Every state change requires computing several hashes).
In graphene you see some places where the opposite decision is made. Graphene's databases could be exposed to a fast scripting environment where users could submit code.
So to answer the question, I think blocktime is almost completely independent, while throughput and cost favor native contracts only by as much as compiled code is slightly faster than advanced VM tech like java