All,
Since I am currently without perl, ikegami was kind enough to provide the following regarding the optree differences:

Given the following !.pl: 
sub goto_scalar { goto $ref;  }
sub goto_amp    { goto &$ref; }


>perl -MO=Concise,goto_scalar !.pl
main::goto_scalar:
4  <1> leavesub[1 ref] K/REFC,1 ->(end)
-     <@> lineseq KP ->4
1        <;> nextstate(main 1 !.pl:1) v ->2
3        <1> goto sKS/1 ->4
-           <1> ex-rv2sv sK/1 ->3
2              <#> gvsv[*ref] s ->3
!.pl syntax OK

>perl -MO=Concise,goto_amp !.pl
main::goto_amp:
7  <1> leavesub[1 ref] K/REFC,1 ->(end)
-     <@> lineseq KP ->7
1        <;> nextstate(main 2 !.pl:2) v ->2
6        <1> goto sKS/1 ->7
5           <1> refgen sK/1 ->6
-              <1> ex-list lKRM ->5
2                 <0> pushmark sRM ->3
4                 <1> rv2cv[t2] lKRM/33 ->5
-                    <1> ex-list lK ->4
-                       <0> ex-pushmark s ->3
-                       <1> ex-rv2cv vKPRM*/1 ->-
-                          <1> ex-rv2sv sK/1 ->-
3                             <#> gvsv[*ref] s ->4
!.pl syntax OK
[download]

At a glance, a masked bandit core hacker suspected the rv2cv extra opcode was necessary but that refgen thinking it needed to boil down a list as though it needed to handle \($a, $b, $c) or some such probably wasn't - though I may have misunderstood him.