The BDC source here are in a MS SQL database. In most of the test the SharePoint indexing server, SharePoint SQL and the BDC data source SQL are in same server (a very powerful one).
Initially the speed was very slow (7.5K per hour) and seems it didn”t downgrade as the crawled records accumulated all the way to 1+ million. Later it turned out to be that the source DB was not properly indexed. So the source DB was the bottleneck (CPU was constantly 90%+).
After the index was added to source DB, the speed became 160K per hour. But as the crawled records went up, the speed was slowed down to 40K per hour (with 2 million records).
The space it uses seems to be about 5 times of the original SQL database.
Small note: it will take about 18 minutes to get ID list of the every 1 million records.