Since the BCM2835 datasheet doesn't exactly specify the set-up time for
the GPIO Pull-up/down Clock Registers there was an assumption of 150 cycles
at a clock rate of 1 MHz. During a discussion [1] in the Raspberry Pi forum
it turns out that clock rate refers to the VPU which has a rate of 250 MHz.
So we can reduce the delay to a sensible value and update the comment above.
I tested this optimization with a Raspberry Pi B and a multimeter.