This is an almost exact twin of my first LED mod (way back in '08, a couple months before I joined CPF), and your results are about like mine. IIRC the initial fall-off was about 20-25% for mine. With better heat-sinking, the drop-off will be less, but what you've got is reasonable. You should also notice the light increases a bit just before it drops out of regulation -- less power burned in the linear regulator = less heat = more efficiency.
You're dumping over 13W into the head, it's gonna get warmer than the "usual" 3D P7 builds. Packing foil around the pill area is a bit iffy -- it's so far down that the foil is hard to keep seated well while assembling it (last thing you need is it shifting and shorting out, don't ask how I know ;)), and the pill does conduct heat pretty well through the threads into the reflector. You've got the cylindrical surface near the front of the reflector, you can wrap foil very tightly around that until it's compressed as it goes in the Mag head, and because it's the last thing in, you can be sure it wound up right where you want it. Also reduces the tendency for the dropin to spin and twist up the wires as you screw the bezel+lens on...
Did you solder the P7 down, or use thermal paste? I soldered mine down, and run the batteries backwards (making the light housing negative -- though if there are no dings in the anodize where the reflector touches, that won't matter.) Complicates the driver setup a bit, though -- if I were to build it again, I don't think I'd bother.
Also, about the donut issue at close range -- you'll always see that in reflector lights, even with single-die emitters. The center portion of the reflector directs no light forward, so the outer parts cast the entire beam. At short range, this leaves a ~1" inside/ ~2" outside donut pattern. At longer range, the 1" and 2" dimensions stay the same, but the light spreads with distance both to the outside and the inside (and past). At arbitrarily large ranges, these reflector characteristics become arbitrarily small compared to the beam spread, and you see the "true" pattern. The P7, being a quad-die, would still have a hole (different effect than the close range one, this is a projection of the gap between the dies) in a perfect reflector, but the OP on that reflector smooths it out nicely and gives you a smooth flat beam -- makes me quite happy.
