Personally, I think Linux should allow binary drivers. Most hardware is useless in a few years anyway, so what good is having the source? Compare that to the OS, where it can live on for decades.
Linux still runs on hardware which is older than Linux. There is nothing inherent in the hardware which causes it to be "useless" after a few years. It is only useless when it is impossible for the individual user to support its use, and this typically happens when the manufacturer loses interest in your platform or when the new (and sometimes improved) generation of hardware shows up.

Unfortunately, the new generations of hardware are not always improved from the end user's standpoint. Modems, network devices, low-end sound, and keyboards are a few examples where manufacturers cripple newer products in order to sell for a lower price and gain volume.

A common tactic is to move hardware functionality into software to save engineering costs and material costs. Frequently, this inconveniences the user even while the product is being supported by the manufacturer (I don't know of anyone who hasn't complained about a Winmodem at some point or another), but thanks to trade secrets and braindead software copyright law, as soon as the manufacturer disappears, the hardware that the user paid money for becomes useless. Not because it's obsolete, or defective, or incapable of performing to specification. It's because the manufacturer didn't care enough to give others the necessary tools to fill in for them, once they decided supporting their products (or staying in business at all) was too expensive for them.

That's not really a fair bargain for individual users who don't throw their single PC away and buy a new one every three years, which is why we would like to discourage the practice of maintaining binary-only drivers in favor of open-source ones. Having to change hardware because the manufacturer's business plans changed *after the date of purchase* is not something that leaves a good taste in my mouth.

You're welcome to disagree, but the vast majority of kernel hackers are taking the opposite position, and they're the ones writing the code.

Regarding software copyrights, I think software authors should be entitled to copyright on their binary code only if they place the corresponding source code in escrow with the Library of Congress. That way, when the company dissolves or the copyright expires (heh), the public actually gets something useful in return, as opposed to rights to an indeterminate binary blob with piles of unfixed bugs and design issues that has to run in a legacy emulation environment or on suboptimal hardware.