but I really don't understand why we have to implement transaction filtering in a way that is invisible to the user.
Becuase setting it sanely is not obvious. There is no NAK in the bitcoin protocol. If you go an set it to 1.0 or whatever you're effectively a 1/8th DOS attack on your peers, if a bunch of people set it to totally random value transactions will start randomly failing because they end up with peers that aren't forwarding the transactions. I explained that in the pull request thread linked to in this thread as well as in some of the other threads here on the subject.
"you should probably not touch this unless major Bitcoin ninjas that you trust are telling everyone to touch it".
True, there is a very strong bias to the status quo. If you are who I think you are then you'll get what I mean when I say that it's a lot like the old days at FPC.
People should stop saying that a billionth of a bitcoin which is 8 places from the decimal, also called a Satoshi, is the smallest unit possible now. That may be strictly true on a purely technical protocol level but if the Qt client and the main mining software is in effect limiting confirmations to a certain size then that essentially screws bitcoin over as a micro currency, especially if it skyrockets in value again.
If it sky rockets in value again the limit can be removed. There are no permanent changes being made. The protocol is not being changed only the software.
There is a certain amount of truth to that but then we get into the design point of view. Why have an arbitrary limit of 8 places on the protocol in the first place, when technically a much larger limit would have been possible, if only to circumvent it later on. I realize that this brings about the idea of an originalist design version adaptation argument but even if Satoshi couldn't foresee this being a problem, and I don't see how they couldn't, I'm not sure this breaks the system to the point where such a change is necessary, much less desireable.