I'm no expert but here is my take...
If you *always* only ever use the MyType implementation of IMyType, then there is no difference (i.e. I believe you will get pretty much the same impementaion using either method). However, the non-register method couples you to that concrete implementation.
This may not be a problem depending how and where you are using IMyType/MyType.
Generally speaking decoupled is a better pattern, leading to less rigid systems. You could, potentially, create a new MyType2 that you use in certain circumstances (customer specific maybe?) so you would register it for those circumstances and MyType otherwise.
One place that decoupling can help is when Unit Testing - an interface is generally easier to Fake.
So my gut feeling is that you are probably right to use the method you are currently using, but that only you/your colleagues can decide if the extra work involved creating and using the interfaces is worth that potential benefit of decoupling.
I'm sure someone else can create a more compelling argument, but that's the best I can do :)