Trezor has the capability to use an alternative entropy source when generating the private key. Not sure whether you can literally seed it yourself (using a physical entropy source), but it can definitely take entropy from the machine/OS you're using the Trezor with.
There are programs to generate seed words from physical dice rolls. You could use these and then initialize the Trezor from the seed words. You would need to use a trusted computer to do this however. If you used a read only operating system and vetted software then you would be OK.
When I first got my Trezor I initialized it and wrote down my seed words. I then sent it a tiny amount of funds. Then I reset my Trezor to the factory settings. Then I booted up an offline computer system and used it to recover the Trezor from the seed words. Then, using my regular computer I connected the Trezor and verified that the funds were there. Next, I took the piece of paper with the seed words and put it in a safe, fireproof place. Finally, I moved the bulk of my funds to the Trezor.
The best part of Trezor is that it's secure, easy to use, and most importanly, easy to use securely. I agree with the people who suggest it makes sense if you have more than 5 to 10 bitcoins that you access more than once a year or so.
Before I got the Trezor I was highly suspicious of the physical security of bitcoins. I used Armory and off-line Electrum wallets, but it became obvious that it was just a matter of time before I committed some serious operational security breech. Plus, it was horribly inconvenient to spend my bitcoins.
Ultimately, there will be trusted computers that can run a mixture of hostile and trusted applications and allow a single TPM chip to protect applications from each other at very low cost. However, given the complexity of such situations and the likelihood that all large scale chip vendors and computer system manufactures are subverted by TPTB, Trezor seems like a very good deal, at least for the next few years.