Hello,
 
perhaps interesting info for some (hardware) developers of gumstix:
I modified (and slightly optimized, e.g. small loop unroll) the pxaregs application (user (!) mode application, direct memory access on registers) for toggling and querying a pin (in my case it was GPIO_62). I measured it on a connex-400g.
 
Following are the amount of accesses on the pin (only rougly measured, just to get a feeling how fast it is...), in which time, and throughput:
 
Reading (loop, 10.000.000 times, each loop 10 read accesses, so total 100.000.000 accesses): 12 seconds -> 8,3 million accesses / second
 
Writing (loop, 10.000.000 times, each loop 10 write accesses, so total 100.000.000 accesses): 15 seconds -> 6,6 million accesses / second
 
Regards,
 
        Martin