Learning to dump: The hard a way

Remember the problem with ARCHIVE engine in the upgrade from 5.0 to 5.1?

Well...I had to mount a parallel MySQL 5 server and restore those tables from the backup. Now the trick is to do a mysqldump...the thing is...what would be the best way to dump the tables (as the avg number of rows is around 20millon)?

The first approach I tried: a bash script that goes table to table doing:
-mysqldump thedb thetable |mysql -htheotherserver

Problems? The dump is going very slow....moreover I learn something the hard way: Mysqldump by default load all the table in memory before dump it....the result? somewhen in the morning the server ran out of memory and crashed......

This week that comes I will try other approaches to the problem....