It's entirely possible my circuit isn't accurate, I built it with the idea in my head and then fiddled a bit with it on veroboard before it worked so the diagram is from memory.
ejoyu your circuit is nice, shifting D4 maybe what my 'fiddling' resulted in, but it certainly looks cleaner than mine.
Looking at the shunting problem you can probably increase R2 as the red LED will require less current than the XM-L and it's forward voltage should be lower (red LEDs normally are although your measurements don't back this up). The other possibility is to switch LEDs so its the rear that's less bright. I get enough comments about how bright my back light it so it can probably be reduced a bit.
I agree with your calcs, I don't see how the Caps can be overcharged as the maximum voltage accross them can only be Vf for the 2 LEDs and 1 1N5818 minus Vf for the 1N4001. That's the beauty of 5.5v caps as for this application.
To jdp's point about seperate rectifiers, I agree it's less confusing but are you getting as much power out of the dynamo? My starting point was two LEDs in series that could take > 500mA. If the dynamo can be considered a constant current source then keeping everything in series gets the most out of it unless I'm missing something?
Last question, what did you use to draw the circuit? much nicer than mine...