A roller chain is the most efficient reduction possible. [but]
The roller chain makes noise. That noise is transferred through the 2 meter high mast to the deck of the boat and a boat is like guitar case.
Roller chain and cog belts are very efficient. But I think V-belts are underappreciated. They have 95-98% efficiency when first installed and if kept properly tensioned, and only drop to about 93% (before they start squeaking annoyingly) if they're allowed to go slack over time. If you have 5% more wind power than you need, you might find the quiet worth taking three extra minutes per hour of mill operation to get to full charge. (And if you're shutting down the mill to get quiet for sleeping, think how much more power you could collect if you didn't have to do that.)
If you do try a v-belt drive, adjust its tension when you install it and again after it's run-in for a few days. After that you should only need to check/adjust it a couple times a year or less often to keep it working efficiently, even if you left the mill up continuously. (Or just tweak it if it starts squeaking.) Like tires they last for years (longer if you give them a sunshade), work in a broad range of weather conditions, and are available inexpensively worldwide in a range of sizes.
In fact the wing always turns at max 
I think You guys call it stalling speed.
Actually the speed is limited by the load voltage plus the charging system voltage drop. As the mill speeds up, the voltage rises. As the voltage rises above the cut-in voltage (battery plus diode drop) the current starts to climb (limited by the excess voltage divided by the winding, wiring, and battery series resistance (approximately, since the diodes aren't ideal so there's a little curve to it)). The current produces a retarding load on the shaft, which keeps the blades from speeding up as much as they would if they were unloaded. Because the current rises rapidly with overvoltage above cutin, the speed while operating is close to a constant related to the system voltage, substantially lower than the free-wheeling speed of the unloaded blades, rather than proportional to the wind speed. (If the diodes were ideal, the wiring and windings were superconductors, and the batteries had no series resistance or other voltage drop from the redox-reduction voltage gap so they were a hard voltage source, etc. the blades would spin up to the cutin speed and run at EXACTLY that voltage-controlled speed regardless of how much higher the wind speed got.)
You see the wind-slip phenomenon especially clearly in water pressure turbines, such as the pelton. Unloaded, the cups ride along with the jet of water. Loaded for maximum energy extraction, they turn the water around and it leaves the cups essentially stopped, with just enough momentum left for it to get out of the way. This leaves the turbine running at almost exactly half the unloaded speed. For wind turbines it's not as simple: They're momentum turbines for a compressible fluid, subject to the phenomena Betz analyzed, and the airfoils' performance varies in a complex fashion with the effective angle-of-attack, as well. Still, RPM under ideal load is in the ballpark of half that of an unloaded rotor in a given wind speed.
Stalling is the phenomenon where the blades are slowed so much relative to the windspeed times unloaded TSR that the wind on the backside of the blades starts to detach on the trailing side - like your sails when you've hauled them in too far while pinching. When that happens they shed vortices and sound like a helicopter (rather than flapping like the trailing edge of your flexible sail). They also lose power substantially. (Some mills use this for furling, though it depends on the electrical load continuing to be present, shakes the mill, makes a LOT of noise, and has other problems, so we prefer things like the gravity or spring based furling tails.)
If you let the mill speed up and use a buck converter to drop the voltage to what's desired (boosting the output current in proportion), you can get more power in higher winds. Max power point controllers do this and adjust the load on the mill so it spins at the speed that gets the most power, but just using a buck converter regulator gives you some of the benefit. (Use a relatively high current one: If it goes into current limiting it may not load the mill enough for the furling to work properly.)
Power available from the mill goes up with the CUBE of the windspeed, while charging current of a straight diode - to - load system goes up with the first power of the excess of the mill's RPM over the cutin speed. So a MPPT controller, or even just a buck regulator, easily pays for its own small losses with a substantial profit at wind speeds even moderately above cutin. (Also: counter-intuitively, just adding some resistance to the simple diode-based circuit can let the mill speed up enough in winds moderately above cutin to provide more charging current to the battery than without the extra resistance, although this hurts you in winds only moderately above cutin, when you need power the most.)
Generated voltage goes up with the RPM, which at ideal loading tracks the wind speed. That means the alternator output current in a max-power-point system goes up with the square of the wind speed. Resistive heating is proportional to square of current and thus goes up with the FOURTH POWER of wind speed. So it's important, with MPPT or approximations like a buck converter voltage regulator, that your furling works correctly, to avoid frying the alternator in high winds.
The Ametek makes zero to 15 volts depending on the wind speed. The diode loses 0.65 volts, that is almost 5%. With a mosfet the gen would start charging 5% earlier.
Right. Lower cutin is great. You also save the losses from the current through the voltage drop, so you continue getting more as things speed up. (It's about like the wind being higher by 5% of the cutin wind speed.)
