Yes, ext2 will do split reads in this case. That is, whenever you attempt to read an 8K database block, ext2 will have to perform two distinct reads. If you had allowed the ext2 filesystem block size to default to 1K, then eight distinct reads would have been needed. However, because ext2 is very good at contiguous allocation and read-ahead, the split reads are not very expensive. Nevertheless, for an OLTP database you will probably notice some gain by switching to a 4K database block size.

Our DB (mainly OLTP) was built with a db_block_size of 8k. The ext2 filesystem (Linux) has a 4k block size. As far as I know, 4k is the maximum block size for ext2. Do you think it's worth rebuilding the DB with 4k blocks in order to match the ext2 block size?