Why is it not good to use recursive inheritance for std::tuple implementations?

Question

In this question, Howard Hinnant said

Some implementations of std::tuple use recursive inheritance. But the good ones don\'t. ;-)

Answer 1

I don't recall Andrei Alexandrescu's GoingNative 2012 talk exactly, but he talked about this point, and one of the points he mentioned was memory layout. If I have a std::tuple, it will be in memory as char, short, int and this layout will take (on my system) 4 bytes more than if they were laid out as int, short, char. R. Martinho Fernandes has reminded me that the best thing to do would be to order these in memory in an order that minimizes padding, which is neither the order given nor reverse order. (Naïve inheritance does reverse order).

If I write std::tuple, a tuple that works by naïve inheritance would put these in the order char, short, int in memory, using 3 bytes of padding, when optimal has zero bytes of padding. (Either int, short, char, char or char, char, short, int).

Assuming I'm right that it's about padding, then R. Martinho Fernandes said "[my argument] doesn't preclude the use of recursive inheritance for the actual implementation in the optimal order.", so that is why I specify that naïve inheritance is bad.

(The order in memory does not mean that get<0> will give a different object, and R. Martinho Fernandes correctly notes that the order should be invisible to the user. However, these were the points as I have been reminded from the GoingNative event.)

The video is at http://channel9.msdn.com/Events/GoingNative/GoingNative-2012/Variadic-Templates-are-Funadic and the slides are at http://ecn.channel9.msdn.com/events/GoingNative12/GN12VariadicTemplatesAreFunadic.pdf.