Theoretically, the CLR-VM should be able to reach better performance than the JVM, since it has value-types. Within the JVM an array of some objects is really an array of references, which is bad for your hardware caches and requires an additional pointer dereference. The question is how significant that is for non-numerical applications.
Theoretically, your sufficiently advanced compiler and/or JVM will figure out where they can replace objects by value types. So, theoretically, that does not matter.
The difference should be visible in practice, though. AFAIK, it is not really visible, though. That probably is because the JVM has been tuned longer.
