That's the risk with making a top-down decision for a system wide TLD instead of letting people choose their own TLDs and taking the risk individually.

The approach I'm going to take (unless I decide I like your project enough to halt development on mine) is to not set a global TLD for this system, but let users register their own TLDs.  The resolver would by default give priority to ICANN TLDs over those from the distributed system (to make the transition smoother for normal people who might not be quick to stop using com/net/org, etc).  Users would be able to change a setting if they wanted to ignore all ICANN TLDs, or choose certain ones to use/ignore.  All of the DNS information for TLDs and sub-domains would be stored in a DHT, and the DNS settings on the user's machine would point to a locally running DNS server which acts as the gateway to the DNS info stored in the DHT.  I'm not going to choose a TLD for everyone else, and TLD conflicts between the system and ICANN can be resolved by the users of the system, without relying on me to fix them.

The decaying network fee idea is interesting...I'm not sure what I think about it, yet.  How many namecoins will the system create?  Is there a set limit like bitcoin (I hope so), or will they be generated forever?  Are the namecoins destroyed by the network fee ever replaced (not that I think they need to be)?