To check this out a bit, I decided to take the supplied supply unit, put a FET (a BUZ102SL) in between to switch it on and off with, and made a debounced signal to put the FET in on and off mode nearly instantly.
I used a 74HCT00 to make a bi-stable circuit, tapped the +5 Volt from the wall wart to feed it, now I can easily switch the Parallella (a 7010 with connectors) on and off, knowing for sure the FET will supply voltage very fast (and only very little loss about 60mV), and completely de-bounced (no switch artifacts).
Works fine, and I might have a higher probability percentage of functioning USB on my Conrad hub, but: still sometimes the USB didn't start up.
Another problem I've been looking at is that I sometimes got file errors at reboot, on my Parallella supplied sd-card, even though most of the time I filled it with the right OS and files, and it first works fine. I've been pretty neat in using "shutdown -H" and waiting till I was pretty sure the ARMs were in Halt mode to pull the plug out of the wall socket. Yet every now and then, of course after having put a lot of new install stuff on the card (like audio drivers and tools and update Adapteva examples), it creates garbage at startup, too much to find a kernel (FPGA program does continue to work, the LED comes on), and when I take out the card to check in my well working card reader on a Linux system, indeed it says it needs repair. Sometimes FSCHK will repair it, but a couple of times (booting) later, and it ends up having to be rewritten from start. Then it works again.
So I tried if a de-bounced "off" of the board will remedy these occasional file system glitches. Just starting X up, login in, doing a few things, and then killing the power abruptly with the new circuit I made worked without visible error, tomorrow I'll try a few more times with a new "shutdown" to see if the voltage drop of pulling power supply out of the wall socket is the cause of the disk corruption problem.
T.