> So nowhere in this scenario is T1 executed again?

"Retry" is *not* among the main topics of the classic STM paper. 

You can let the initiator of $T_1$ call `StartTransaction()` upon each `Failure` until it receives a `Success`. The authors have just ignored it probably because they want to focus on the ingenious STM concept.  

Retry is of course a big topic itself, for instance, in [Composable Memory Transactions](http://research.microsoft.com/pubs/67418/2005-ppopp-composable.pdf).