Do you think that this is strong evidence? What does "most" of the information mean? How would you design an experiment to test infinite memory capacity?

One of the classic ways to demonstrate that a lot of knowledge can be retrieved is to retrieve from a person the number of steps in the stairway of the house in which he lived as a child.

I personally have IRC and ICQ and MSN messanger transcripts that go a long time back with contents that you could potentially retrieve.

The fact that you get in some savants total memory recall of some particular type after damaging their brain is also good evidence.

(I think wedrifid's "physics says otherwise" makes this discussion rather pointless.)

As far as physics is concered maybe a human being that's 1,000,000 years old runs into problems with storing his memories. That doesn't mean that an issue for human operating in todays world.

Keeping memories stored requires energy, how does it make evolutionary sense to store memories you never recall?

Because the main evolutionary reason that we store information in our brain isn't to recall memories. It's to pattern match what we experience into categories and make decisions based on those categories. For pattern matching it's useful to keep storing all information but unnecessary to retrieve individual instances of memories.