(Though, I half expect Perl to do @in= @in[@idx]; in-place.)

Seems to me from a quick look at pp_hot.c in perl.git that aassign ops where the LHS and RHS contain common members is actually slower, not faster.

>perl -MO=Concise -e"@in=@in[@idx]
e  <@> leave[1 ref] vKP/REFC ->(end)
1     <0> enter ->2
2     <;> nextstate(main 1 -e:1) v ->3
d     <2> aassign[t7] vKS/COMMON ->e        <-- Note "COMMON"
-        <1> ex-list lK ->a
3           <0> pushmark s ->4
9           <@> aslice lK ->a
4              <0> pushmark s ->5
6              <1> rv2av[t6] lK/1 ->7
5                 <#> gv[*idx] s ->6
8              <1> rv2av[t4] sKR/1 ->9
7                 <#> gv[*in] s ->8
-        <1> ex-list lK ->d
a           <0> pushmark s ->b
c           <1> rv2av[t2] lKRM*/1 ->d
b              <#> gv[*in] s ->c
[download]

         1         995:     /* If there's a common identifier on both 
+sides we have to take
         1         996:      * special care that assigning the identif
+ier on the left doesn't
         1         997:      * clobber a value on the right that's use
+d later in the list.
         1         998:      */
      5492 <<      999:     if (PL_op->op_private & (OPpASSIGN_COMMON)
+) {
      2062 <<     1000:     EXTEND_MORTAL(lastrelem - firstrelem + 1);
      5492 <<     1001:     for (relem = firstrelem; relem <= lastrele
+m; relem++) {
      5540 <<     1002:         if ((sv = *relem)) {
        18 << br  1003:         TAINT_NOT;    /* Each item is independ
+ent */
      5492 <<     1004:         *relem = sv_mortalcopy(sv);
        18 << br  1005:         }
      5492 <<     1006:     }
         1        1007:     }
         1        1008:
[download]