Here is the program:

zip_filter :: (Num a, Ord a) => a -> [a] -> [a] -> [a]
zip_filter x as bs = zipWith (+) (filter (<x) as) (filter (<x) bs)

GHC generates this:

poly_z_smp = \ (@ a_aiz) _ -> GHC.Types.[] @ a_aiz

T.zip_filter =
  \ (@ a_aiz) $dNum_ajm $dOrd_ajn eta_B3 eta_B2 eta_B1 ->
    letrec {
      go_smr :: [a_aiz] -> [a_aiz] -> [a_aiz]
      go_smr =
        \ (ds_ak5 :: [a_aiz]) ->
          case ds_ak5 of _ {
            [] -> poly_z_smp @ a_aiz;
            : y_aka ys_akb ->
              let {
                r_smt :: [a_aiz] -> [a_aiz]
                r_smt = go_smr ys_akb } in
              case GHC.Classes.< @ a_aiz $dOrd_ajn y_aka eta_B3 of _ {
                GHC.Bool.False -> r_smt;
                GHC.Bool.True ->
                  \ (ds_alR :: [a_aiz]) ->
                    case ds_alR of _ {
                      [] -> GHC.Types.[] @ a_aiz;
                      : y_alW ys_alX ->
                        GHC.Types.:
                          @ a_aiz (GHC.Num.+ @ a_aiz $dNum_ajm y_aka y_alW) (r_smt ys_alX)
                    }
              }
          }; } in
    go_smr
      eta_B2
      (GHC.List.filter
         @ a_aiz
         (\ (ds_djz :: a_aiz) ->
            GHC.Classes.< @ a_aiz $dOrd_ajn ds_djz eta_B3)
         eta_B1)

Eta-expanding go_smr would result in much better code.