It can never happen that Dapper is much faster than ADO.Net. Period!
We know that Dapper team has written an advance mechanism via IL to just speed-up the process, but still it is running on top of ADO.Net. They are only abstracting the ADO.Net and processed everything on top of it by converting the actual data (rows) back to the object models (POCO or DTO), and that would add additional time doing it.
I think everyone must understand the separation of concern of the ADO.Net layer from any other ORM available on this world.
Or else or unless, the Dapper team will write their own version of DbDataReader, then, there could be a possibility that they are much faster, but never ever it will happen until they do it.
With EF, there is no question that Dapper is much faster than it as they are the most lightweight ORM right now that is available.