Well, just a comment as a desktop user that had no trouble reproducing this problem:
I have a usb stick which has poor write speed, which I periodicly copy ~16gb of linear data to. If I just left the machine alone, it would struggle along at a rate of perhaps 1mb/s - 3mb/s, with occasional pauses where almost no data moved. If, however I were to try and use a browser, not only would the browser hang for upto minutes, but other applications would hang, not updating their windows, AND the copy process really went to hell, trickling out a few kb/s and requiring hours to complete. Oddly, the only way I could get things going again after it entered this IO hell state, was to just cat gigs of disk files. (this just got the copy going again-- no help for hanging applications.)
This seems the definition of unacceptable behaviour on what is an otherwise capable dual core amd64 box with 4gb of ram. Eventually, after trying many things to tune swap/page and even formating the USB sticks to start their data clusters on 128k clusters, etc. (before that, the IO was even slower...) I finally just rebuilt the kernel without transparent huge pages. I wish I had heard they might be at fault earlier, but for me that was just a guess.
So, unless some fix is implemented, it would seem to me that transparent huge pages is simply an option that should at least be documented as being potentially completely destructive to desktop usage.