Order of Mia Sets

JavaScript (ES6), 162 bytes

(a,b,g=a=>a.map(n=>e[n]=e[n]+1||1,e=[1])&&[[...e].every(n=>n==1),...e.filter(i=x=>x).sort(h=(a,b)=>b-a),...a.sort(h)],c=g(a),d=g(b))=>d.map((n,i)=>n-c[i]).find(i)

Explanation: Takes two arrays as parameters. g converts each array into a list of counts. The list is then checked to see whether it corresponds to a set 1..n. The counts are sorted and the sorted values are concatenated. The two results are then compared. The return value is a positive integer if the second array wins and a negative integer if the first array wins, otherwise the falsy JavaScript value undefined is returned.

Jelly, 16 bytes

ṢŒrUṢṚZ
Ṣ⁼J;ǵÐṀ

Takes a list of lists each of which represents a roll (so can be more than two if wanted) and returns a list of the winner(s).

Try it online! ...alternatively here is a version which sorts the rolls from weakest to strongest instead.

How?

Ṣ⁼J;ǵÐṀ - Main link: list of list of dice rolls, L
     µÐṀ - filter keep maximal (i.e. sort L by the previous link as a key and keep maximums)
         -                                            e.g. [5,3,1,3]
Ṣ        -     sort roll                                   [1,3,3,5]
  J      -     range(length(roll))                         [1,2,3,4]
 ⁼       -     equal? [1,2,3,...n] beats everything        0
    Ç    -     call last link as a monad with input roll   [[2,1,1],[3,5,1]]
   ;     -     concatenate                                 [0,[2,1,1],[3,5,1]]

ṢŒrUṢṚZ - Link 1, rest of sort key: dice rolls        e.g. [5,3,1,3]
Ṣ       - sort the roll                                    [1,3,3,5]
 Œr     - run length encode                                [[1,1],[3,2],[5,1]]
   U    - upend (reverse each)                             [[1,1],[2,3],[1,5]]
    Ṣ   - sort                                             [[1,1],[1,5],[2,3]]
     Ṛ  - reverse                                          [[2,3],[1,5],[1,1]]
      Z - transpose                                        [[2,1,1],[3,5,1]]
        -     ...this is a list of: 1) the group sizes descending; and
                 2) the face values of each group, descending across equal group sizes