Yes, you can easily use Kodo in conjunction with a server farm or cluster, with a load balancer in front of the farm. Kodo provides support for synchronizing its caches across JVMs, so you can use Kodo's data cache when in a clustered environment as well.

Yes. By default, Kodo does not hold hard references to objects read from the database, so, provided that you do not hold hard references to more objects than you can fit into memory, you will not run out of memory when iterating through large data sets.