How optimize new RTTI (in XE4) ??? it's very slow !

How optimize new RTTI (in XE4) ??? it's very slow ! 
when I use the old rtti method (in TypInfo, Set/GetxxxProp family...) it's 5/6 times faster !  mostly is the TValue. Why ?

I writed a function in a class to scan all published properties (with rtti) and compute CRC 32bits on it ; my version with old rtti is 8 times faster than the new rtti !  unbelievable !!!

Unfortunately , for this example, I don't have the new rrti version (because too slow). but the old rtti (and the faster) is:

function ComputeValueCRC(AOldCRC:Cardinal; const AObj:TObject; AProp:PPropInfo):Cardinal;inline;
var
  s:string;
  ws:WideString;
  i:Integer;
  i64:int64;
  f:Extended;
begin
  case AProp^.PropType^^.Kind of
    tkInteger, tkChar, tkEnumeration, tkWChar, tkSet:
    begin
      i := GetOrdProp(AObj, AProp);
      Result := ShaCrcCompute(AOldCRC,@i, Sizeof(i));
    end;
    tkString, tkLString, tkUString:
    begin
      s := GetStrProp(AObj, AProp);
      Result := ShaCrcCompute(AOldCRC, Pointer(s), s.Length * Sizeof(char));
    end;
    tkInt64:
    begin
      i64 := GetInt64Prop(AObj, AProp);
      Result := ShaCrcCompute(AOldCRC,@i64, Sizeof(i64));
    end;
    tkWString:
    begin
      ws := GetWideStrProp(AObj, AProp);
      Result := ShaCrcCompute(AOldCRC, Pointer(ws), Length(ws) * Sizeof(wchar));
    end;
    tkVariant:
    begin
      s := VarToStr(GetVariantProp(AObj, AProp));
      Result := ShaCrcCompute(AOldCRC, Pointer(s), s.Length * Sizeof(char));
    end;
    tkFloat:
    begin
      f := GetFloatProp(AObj, AProp);
      Result := ShaCrcCompute(AOldCRC,@f, Sizeof(f));
    end;
    else Result := AOldCRC;
  end;
end;

function TObjectHelper.TUserDataObjProps.ComputeDataObjectCRC(
  AObj: TObject): Cardinal;
var
  Prop:PPropInfo;
begin
  if not FCrcInit then InitCRC;
  Result := cCRCINIT;
  for Prop in FCRCProperties do
    Result := ComputeValueCRC(Result, AObj, Prop);
  Result := not Result;
end;