I've been testing [iSCSI](http://en.wikipedia.org/wiki/ISCSI) performance for the past few days using an iSCSI volume on a [ReadyNAS Pro](http://www.readynas.com/?cat=24) (a 6-bay dual-gigabit NAS box). I've got the ReadyNAS Pro configured with 6 drives (of varying capacities between 1TB and 2TB each), and am running a dedicated gigabit line between the 2nd port on my Mac Pro and the 2nd port on the ReadyNAS Pro (bypassing routers/switches altogether). This is the preferred method for connecting to iSCSI targets.
Read speed is as expected, ramping up until it maxes out at around 102 MB/s (megabytes) for large block sizes. Write speed, however, was all over the place. The iSCSI target would time out for periods in which there was no traffic going over the network, resulting in inconsistent performance. Running the benchmark multiple times resulted in totally different write results.
Note that this is a real world test scenario. I have 5 machines connected at all times to the ReadyNAS Pro and have Crashplan backing data up to it constantly, as well as multiple cron jobs rsyncing data from web servers. This may be why write performance is so inconsistent, and provides a good argument for using a dedicated iSCSI box for primary storage.