When carrying out 64MB reads with 32 threads, the latency was very similar between XFS, EXT3, EXT4, and Btrfs. EXT3 had a latency of merely 13 microseconds while XFS and EXT4 were just one microsecond slower. With this test profile, NILFS2 has a rather significant problem as its average latency was 461 microseconds.
A similar issue with NILFS2 was also spotted when it came to 64MB random reads using the Threaded I/O Tester. XFS, EXT3, and EXT4 had average latencies of five microseconds and Btrfs came in fourth with a time of 26 microseconds. With a glacially slow, last place finish was NILFS2 with a time of 919 microseconds.
When increasing the read size from 64MB to 256MB, but sticking with 32 tiotest threads, NILFS2 remained the slowest but by a much more reasonable margin. XFS was the fastest here followed by EXT4 and just behind that was EXT3 and then Btrfs.