I agree with the support of the keyword argument init. However, there is room for debate about compatibility.
If we support multiple input colors, (as the current mapc supports,) there will be some confusion. For example:

mapreducec((a, b) -> max(a, b), *, Gray(1), Gray(0)) # Does `Gray(1)` mean the `v0`?

I don't think this is much of a problem in the current situation. However, issue #207 needs to be considered more carefully.
If the current formats are to be deprecated, we need to prepare for depwarn.

BTW, this isn't an essential issue, but the default value of init seems to depend on the operator op.

julia> mapreduce(identity, +, (1.0, 0.5, 0.25))
1.75

julia> mapreduce(identity, +, (1.0, 0.5, 0.25), init=0)
1.75

julia> mapreduce(identity, *, (1.0, 0.5, 0.25))
0.125

julia> mapreduce(identity, *, (1.0, 0.5, 0.25), init=1)
0.125

I also set one of the milestones for v0.13 to be the provision of iterator interfaces (cf. #131 (comment))
In that case, reducec and mapreducec may not be necessary anymore.

julia> components(c::Color3) = (comp1(c), comp2(c), comp3(c)); # substitute a tuple for an iterator object

julia> mapreduce(identity, *, components(RGB(1.0, 0.5, 0.25)))
0.125

julia> mapreducec(identity, *, 1.0, RGB(1.0, 0.5, 0.25))
0.125

Even if we may deprecate reducec and mapreducec in the future, I think it would be beneficial to change their implementation to reduce/mapreduce-based (unless there is a significant performance degradation).

The ability to omit the initial value is more attractive than the ability to explicitly specify the init keyword argument. Shall we keep the current formats and add support for the new formats in ColorTypes v0.12?