Thanks for clarifying, manganese!
So essentially the only difference with the current system here is the presence of a grace period (which is actually nothing more than longer phase length) and automatic pausing game (if the game is set this way) on NMR.
I'm afraid if we implicate this approach, either most of the games will be started without pause-on-NMR feature or at least 50% games will be paused with a very slim chance to unpause.
I still think that the main problem we should address is demotivation to CD and motivation to take over the CD’s. Demotivation to CD can be achieved by explicitly stating “reliability rating” of each player and adding the feature which allows restriction of joining the games based on this rating (importantly, “reliability rating” is not 100% clean by default – when the new player joins the site he first should build up his rating by participating in certain number of games without single CD; otherwise there would be a loophole to purge your record by creating new account when the old one is shouting “Unreliable!!!”). As to motivation to take over the CD’ed positions, I see three ways for that:
1) separation in the player’s record of the results achieved in the games played from the very beginning from those which come from the games where the player took over CD;
2) ability to improve “reliability rating” as a result of CD takeovers (you can redeem you “CD sins” (“CDins” :)) by doing the right thing);
3) point incentives to take over CD (say, cost of takeover is half of the actual position price, or even some fixed amount, for example 5

).