They don't do well on a sailboat that is moving around in the sun in more or less random ways.
Tim Nolan would disagree with you. however, he has since updated his website and the 4 channel mppt unit he built for his sail boat back in 2002-3 which worked quite well is no longer documented, but he has put together an arduino package that uses as far as i can remember, very similar code, and is quite fast.
There is nothing wrong with hill climbing algorithms. What is wrong (in the context of we've got 50 mips, why not use them) is using constant deviation.
there is a circuit floating around the net that uses a 4066 chip an opamp and a pwm controller to make a rudimentary modulation and demodulation circuit that actually works, but it has no provision to wrap an automatic gain control around the deviation oscillator, so it will always push the PWM around at 50hz, or whatever you set it too. I suppose such a system would inherently be unstable though if the gain is too high, and difficult to build.
However, anyone can look at the PV chart and see that a 2% change in current isn't really that bad, and given that its really +/-1% in either direction from the peak, its really not worth the effort to use active methods to reduce that ripple.
That said, analogue methods will be able to lock on to a secondary MPPT point and can't tell the difference, if these points exist at all (several manufacturers claim their digital processing methods will find the highest mppt point -like ripples in the pass band)
As far as the speed, in theory it can be done very quickly, limited only by the RC time constant of the solar panel themselves.
In practice the fundamental limitation is response time of the dc-dc converter and processing power.
An efficient synchronous buck converter running at 100Khz will be able to make fairly clean audio if you were to drive speakers with it, so we're essentially limited by the processing power alone.
