I'm following the performance test that Dapper community created.
Currently, I'm getting the following after run the tests 10000 times:
I need that something help me to understand why EF is faster. What did I do wrong?
The code that I'm using is on Github
Firstly, EF did definitely get faster: they invested a lot of time improving perf - by most of my measures, it still has noticeable overhead. Consider: EF sits on top of ADO.NET. If EF is reporting results that are significantly faster than ADO.NET, then (and I think you already suspect this): you aren't measuring what you think you are measuring.
I suspect the key issue here is that you are performing all the operations on the same data-context. This means you are probably now just stressing the identity manager and cache. I can't run your code directly (because it uses Oracle etc), but it would be interesting to view the actual queries that are performed. In real world usage, each data-context has a finite unit-of work lifetime. It would be incredibly rare to execute 10,000 operations on the same data-context. More typical would be a handful of operations.
However, unrelated: I have to say: I'm very pleased with your results showing dapper as being within 0.6% of raw ADO.NET performance. This is basically exactly what we were going for with dapper: the full performance of raw ADO.NET, but simply taking away the coding pain.