I was combining 14GB and 3GB partitions. Table type is MyISAM. The thing that looked odd to me is that the CPU was maxed out the entire time and there was negligible IOWAIT. I've also seen a few other issues:
1. It's much faster to create a new, empty table that is partitioned and insert data than it is to create a partition layout for an existing table.
2. Killing a REORGANIZE PARTITION has no effect except for changing the process to a killed state. Meaning that the reorg still completes normally and is also replicated.
We are running MySQL 5.1.52-log on 64-bit Centos 5.6. The machine is an AWS m2.xlarge instance type (17.1GB RAM). Kernel is 2.6.21.7-2.fc8xen. Here is the buffers info. I should note that is was the data copy phase that was slow and not the index rebuild.
bulk_insert_buffer_size 8388608
join_buffer_size 131072
key_buffer_size 7516192768
myisam_sort_buffer_size 2147483648
net_buffer_length 16384
preload_buffer_size 32768
read_buffer_size 2097152
read_rnd_buffer_size 8388608
sort_buffer_size 256000
sql_buffer_result OFF