I tried the slice method and ended up projecting artifacts. Then, tried dedoming and didn’t like what it did to the to the brightness and tint. Thought I’d take a shot at reshaping the dome. The stock dome is shaped to direct light forward which is great for an aspheric. By flattening the top and rounding the sides, the light gets focused to the sides. This should help a lot with reflector lights especially short ones. I’m also hoping it’ll add yellow to the purplish rosy tint of this XML2 U3 2A. I’ll post results later. 
I used a razor blade to get the desired shape, then used 1000 grit sandpaper to smooth. On the right, more phosphor appears at wide angles and less at narrow angles.
UPDATE:
When I get my drivers, I’ll try different refectors and adjust the focus but here’s a rough comparison in grayscale.
Light distribution of mule aimed up:
Focus is only close as I’m holding a standard C8 reflector against the LEDs by hand. The modified spot is actually smaller than it appears. The large corona obscures it:
The LED is focused mostly at bottom of the reflector so it has a large corona. The top half has almost none focused on it which gives it a tight spot. I was shooting for more in the spot so on the next one, Ill take less off the top.
UPDATE:
Didn’t feel like waiting so I installed my crippled Qlite in an extra C12 pill and tried different reflectors. Nothing much changed. I ended up liking the OP best. It gives a beautifully seamless transition from center to spill.
Here’s the best untouched color picture I could get. Both have smooth reflectors:
UPDATE:
You can see how the phosphor image starts to divide where the curvature is slightly too sharp. It doesn’t create an artifact because it’s barely divided and only at a specific viewing angle (in the corona). With a cylinder shape, two distinct images of the phosphor would be seen at nearly all viewing angles.
Here’s a rough estimate of the light distribution from the center.
I don’t see a noticeable tint shift in the spot but corona and spill are a less blue. It’s performing about how I expected. Where the stock dome had a well defined spot and a somewhat defined corona, the modified dome has a small bright center blending into the corona which gradually spreads out into most of the spill like a starburst.
UPDATE:
I think I got it figured out. I shaved a little off the top of an old XML and undercut the base. As a mule, it now spreads about 160 degrees instead of 180. The corona is no longer projected from the bottom of the dome but instead from the top so it’s smaller and not yellow! I think it could be much smaller or even gone completely if I didn’t flatten the top. The spot is bigger because the yellow light that normally makes up the corona is projected higher in the reflector and contributing to it. On the next one, I’ll just make the undercut and leave the top and see if the corona disappears and the spot size changes.
Ignore the brightness differences, the old modded XML is running on less than 3 volts. Note the lack of yellow in the corona.
UPDATE:
It took me a while to get a nice uniform shape but here’s the spherical dome. I found that the bottom needs to be perfectly scratch free to refract the light cleanly upwards. I can’t see how close the wires are so I’m not going any further. Spot size is the same as stock and corona is thinner and doesn’t have that blocky shape. Tint seems a little warmer but my camera exaggerates it.
