Global IO manager - improve auto thread allocation for hyper threading processors.

Global IO manager implementation calls the sysconf function to get the number of online CPU and uses this number in calculating max number of threads to spawn. The number returned by the sysconf is the logical number of cpus which in the case of processors supporting hyper threading returns two cpus per each physical core. There are some noted drawbacks of using hyper threading in some applications. Has any benchmarking or analysis been done to see whether using physical cpu count or logical cpu count in determining number of IO threads provides better performance for our application?