Scott, I understand your position...the box is bonded, and conducts, so there's the electrical continuity you need....and why should we ask more of this connection than we ask of every coupling in the conduit? I will also admit that I am completely unaware of any bad things happening as a result of using the mounting screws as your bond.
Here is where I find the requirement....you are required to have a bond sufficient for anticipated overcurrents, so as to clear a fault (as the code explains). With a ground bar, you need a bond ing connection able to carry everything that bar can- and with multiple circuits, I question the capacity of the screw connections. The code constantly refers us to table 250.66- and the smallest wire listed is #8 copper. Finally, in 250.92(B) the code discusses "bonding at the service."
250.92(B) lista any number of permitted methods of bonding, but using the monting screws isn't one of them. The closest suggested way is #4, which suggests 'bushings with bonding jumpers.'
Another point that has influenced my thinking was demonstrated when I wired an audio recording studio. There I learned that even properly made-up pipe produces some radio static from tiny amounts of arcing at every connection. To reduce this electronic noise, ground wires were run to isolated devices. (The conduit then acted as a "shield" for the ground wires). In using such a jumper, I am trying to reduce the current that will flow through the mounting connections.
Is it overkill? Probably. Does Code require such a jumper? I think so. Has anyone ever been cited? I doubt it.
And, as to the "Panel or sub-panel" issue, I am only speaking of bonding one ground bar to the original ground bar. The connection to the "neutral" system hasn't been changed at all.
