Highly-parallel F# program shows poor CPU utilization

Question

One of the promises of pure functional programming is that it parallelizes well. I'm testing this claim using a F# application with mediocre results. My program runs a large number of MiniMax searches in parallel via Array.Parallel. The MiniMax algorithm is pure functional code - no shared state, no locks, but highly recursive with lots of values being created and destroyed as it searches the tree. There is no I/O at all - everything is in memory. Each MiniMax search takes 5-60 seconds and I'm running about 100 of them in parallel on a fast box with 8 CPU cores. Sadly, CPU utilization peaks at about 65% and is usually in the 45-60% range.

I profiled my app using the Visual Studio Concurrency Visualizer and found that it is blocked about 40% of the time. All of the blocking calls seem to be in the .NET garbage collector or other .NET memory management routines. Is there some way to optimize this behavior without rewriting the entire program in a lower-level language such as C++? It seems clear that the problem is that I'm creating and destroying too many objects, but this is hard to avoid in idiomatic F# code. Perhaps I'm missing some other cause of the synchronization issues?

Thanks.

Update: I made two changes: Disabled hyperthreading and used gcServer in my config file. This dropped the execution time of my test case from 32 to 13 seconds! CPU utilization is also much higher. Thanks to everyone who made suggestions.