I'd looked at cavities at first, but was daunted by "how to get ions into the tank" from one, as well as the lower fields per watt usually obtained. It's illuminating to look at Terman pg 273 as he shows field evolution as a 1/4 wave stub is evolved into a cylinder cavity. That real high concentration near the hot end was what got me going on the stub design, where I could run the gas tube right through that spot and then into the tank via a simple tubing compression coupler. There were some significant tradeoffs here. I couldn't increase the diameter of the stub design without risking "monkey modes" in the large space, rather than transmission line behavior, and I couldn't make the holes for the ion pipe any bigger as they'd become significant microwave leaks. So that's where I wound up for that one -- as is it's pushing some limits, it's kind of too big to be a good transmission line, almost a quarter wave in diameter! In the new one, the issues are going to be more like sputtering, since there's no way to have an electrode-free area where the ions and fields will be and still put it where I want it, and I'll have to use some sort of low-sputtering, high temperature material for the screens (probably Ta wire for the first go, since I have some).
I had another scheme for fusion in mind, that would have required chopping and bunching ions while they were going "slow", and was looking at that klystron type of design you mention to do that. But the doggone heavy ions have such slow transit times at reasonable energies I couldn't make it fly even on paper with reasonable sized stuff. You'd need so much power working over such a short gap you sorta couldn't make what I had in mind at the time.
I'll be making and trying the new design fairly soon, pretty much as soon as I complete the turbo driver design -- it's close now, just fighting some noise issues in the switcher I use to replace the nasty high power series resistor in the original. I might have some fried parts in the proto, so...gotta do some more work there. The additional software needed on top of the opsys I have posted up here is pretty trivial, and will give some nice advantages in control and data acquisition.