I haven't heard an argument against it. To be clear it would still be possible to control witnesses by spliting stake, but it would be much more difficult.

Yes, the only argument against it that I'm aware of is that someone could theoretically get around it by splitting their stake, so it's not a comprehensive solution to any problems.

What are your thoughts? Is our current DPOS system fine? How would you improve it?

I'm not sure exactly how I would improve it, but I would like to see multiple ideas considered and debated, in an effort to find things that work better for the long term.

One thing I am curious about is if downvotes for witnesses would be helpful. If, rather than voting in friendly witnesses Steemit had been able to downvote hostile witnesses I think I would have been more comfortable.

Getting a bit more complex, I also wonder if there could be a separate "ratification" step for which code versions are valid to run. For example, people vote for code version numbers like "23.0" or "23.1" independently of voting for witnesses, and if a witness runs code that hasn't been sufficiently upvoted by people then vote quantity giving them their position gradually decays. This would let witnesses run emergency code when necessary, but would encourage them to get buy-in on code changes before new versions. I haven't thought through all the implications of this, though, so there may be flaws I haven't considered.

One of the complaints I used to see occasionally was that "stale" witness votes would hang around, e.g. if someone does their vote may no longer be expected to track which witnesses are or aren't trustworthy.