In our testing, jumbo frames only provided about a 4% performance increase which did not offset the effort required to put it in place on our storage VLAN. This is with UCS, NetApp and Nexus 5548s running at 10GbE end to end. Additionally, with jumbo frames in place, we have to undertake additional troubleshooting steps whenever an issue arises, to ensure we're not getting fragmented or getting discards as a result of any inconsistency in jumbo frame configuration within the environment.
YMMV, and with less capable switches, one might see much better performance gains especially if the switch ASICs or CPU are not up to par. In an environment utilizing store-and-forward switching, jumbo frames would likely introduce increased latency that offsets or negates any benefit the increased payload might have provided. I frequently saw this years ago while working for an iSCSI storage vendor that did switch validation and compatibility matrices.
There's many variables that will influence whether jumbo frames are beneficial in your environment, and it really warrants testing to decide if it's best for you.
