@Carl Ohai - When I run comps, I don't comparable manufactured homes with stick-built. Whenever possible, run your comps with the same type of construction.
Comps are not an exact science, there's a bit of wiggle room to them, and you want to reserve that wiggle room for the right things. In general, I run my comps within a 2 mile radius (wider for rural), similar lot size (+/- 20%), similar square footage (+/- 20%), within the same building type (condos / residential / manufactured...), and with a similar age range (compare new to new, post 1978 with like, and +/- a decade before that). Ideally I like to have 10-20 properties to compare from, and refine based on condition and actual similarity to the property at that point.
If there aren't enough comps in the area, consider what is in that neighborhood and broaden parameters accordingly. I also look at Active, Pending, and Sold (as separate subcategories in the comp) to see where they market thinks it is, where the market bites, and where the market actually is. All of these will give you valuable information. You can also take this information to fill in any categories that are light (for example, if there are more pending manufactured with days on market higher than the median DOM, that tells you there's some market saturation and to adjust comp accordingly).
Last note, I don't do comps based on # of bedrooms. You can, and this is where the art of comps plays in for me. A 2000 SF house with 3 bedrooms versus a 2000 SF house with 4 bedrooms in my market will comp similarly. The 3 bedroom may have an office, or the 4 bedrooms might feel "tight". Age, overall living space and lot size within an appropriate radius will give you a lot of great info.