Haskell, 495 442 401

import Data.List
d=1:1:1%2
f=filter
p 0="0"
p 1="1"
p n=p(div n 2)++p(mod n 2)
l=length
u z[a,b]=sort.head.dropWhile((<b).l)$m(nub.foldl1 intersect.y(tail.p$31-a).(`m`[d,f(v.group.sort.p)[1..],z#1,y(z>>=p)z]).take)z
w=(=='0')
v x=l x==1||all(all w.tail.p.l)x
y x=m snd.f(w.fst).zip x
x#n=n`take`x++drop(n+n+1)x#(n+2)
n%m=d!!(m-d!!n)+d!!(m-d!!(n-1)):m%(m+1)
main=interact$show.u[0..].m read.words
m=map

It performs reasonably well. Here are a couple of OP's examples:

Flonk@home:~>echo 15 10 | codegolf
[10,23,40,57,58,139,147,149,212,228]
Flonk@home:~>echo 9 12 | codegolf
[3,8,10,14,19,20,21,23,24,31,37,40]
Flonk@home:~>echo 2 10 | codegolf
[0,3,4,5,10,11,12,13,14,21]
Flonk@home:~>echo 12 6 | codegolf
[2,4,5,6,9,10]