There's a provocative title for you. We are attempting the impossible, so either this guy has to be wrong, or we have to be more right than he, and more subtle/clever.
A longer version (his thesis) is here. Warning, it's 18 megabytes. Don't you hate people who just scan in PDF's so they are a long series of images instead of the much more compact original form?
Reading these, as ChrisB and I both have, I think leads well into a discussion of why we think our approaches aren't silly -- we both address the issues described here in our own ways, and I'm convinced that any successful other plan will have to also. I believe that someone calculated that for a straight Farnsworth fusor you need recirculation on the order of 7000 times before an ion is lost to get to gain. I'm sure not seeing that in any tests here, or even close, the time that would take to die off after an ion source would show easily on my gear, and I see very fast shutdown -- perhaps one or two transit times of the ions at the expected energies (and I have hints that the actual energies are much less than expected from oversimple analysis).
However, while this guy is good, he's not perfect either, and not everything he says is quite right. Not that his errors necessarily work in our favor, but it's encouraging that even an amateur can also pick some of his points apart.
For example, to get the ball rolling, Fig 1-1 on page 8 of the attachment is somewhere between wrong and dead wrong, as it ignores the charge on the ions affecting the potential well, and in actual conditions, it may cancel it completely! So we begin right off with a failure to take the reality of the real system into account (not uncommon for academics). Further, the potential well inside a grid is not like that created by a point electrode in the middle -- it has a more complex shape than that. It's not just me saying that one -- many schemes try to keep some electrons around to prevent this (think plain old plasmas) which then become tokomak-like things, which then fall afoul of thermalization, or the spreading out of the ion energy/motion vectors into all degrees of freedom, which kinda wastes most of the input energy on that.
I'll have some more as I re-read these (good to go back once in awhile) but here I'm just getting a ball rolling so we can have "many eyes" looking at this. Large as it is, it's the most compact and complete list of the issues we have in the library (I think).
So have at, people -- this is the real thing we're trying to learn to overcome with various schemes, the things our inventions have to handle in one way or another if we want to have hope of gain -- or even a very nifty neutron source capable of breeding fission fuel. Actually, it left out a few things....but we can get to those in time. Just one example here. Collision properties are assumed smeared out statistically (mean free paths assuming random spatial distribution of the individual nuclei -- and they're not random as their charge tries to space them out more evenly), but in a focus situation they're not, as the density isn't "thermal" and spread out for say, the case of firing one single nucleus at another with accurate charged particle optics. At that point, the probably of collision can be raised with the only limit being the precision of your "aim" and timing, and can go to near 100%. The assumption that probability is purely density in a random distribution of "aims" isn't questioned in these writings either.