Good discussion. David, that is a LOT of records!

One note on the expiration period. I can see why with x billion rows you want to shorten it, but for less extreme environments I've set it to 96 hours, allowing me to survive a replication stoppage over a 3 day weekend without having all my pubs expire. I agree that monitoring should be in place as well, but this saved me some pain a couple times.

Early in my SQL days I had 250 db's replicating into one distribution db. Ran into a lot of contention, ended up solving it by not running all the logreaders at once, didn't think to try the multiple distribution model. The one advantage of my approach (was on SQL 2000) was saving quite a bit of memory by only running a handful of logreaders at once.