The first part of this question has been solved (see comments).

In the book "computational complexity" by Papadimitriou, a Universal Turing Machine is given. But this machine is not concrete, in the sense that it contains variable parameters. For example, the states of the input Turing Machine are represented as binary strings of fixed length. Hence, no matter how long we let this length be, there will always be Turing Machines with so many states that they cannot be represented. Can this machine then really be called a UTM, or at least a template for a UTM?

I also read that a 2-state, 3-symbol UTM has been discovered. This would mean that the action table contains only 6 entries. Am I right in my conjecture that such a machine must have a very complex encoding of the input TM?