Are you running on Windows or Linux or Mac OSX ?
For Windows, it would help if you could show the information of the Task Manager for the running bitcoind program.
For Linux, the information of (the program) top for the running bitcoind.
For Mac OSX, the information Activity Monitor is showing.

This could be a start to try to explain what's going on.

A develop will need to run bitcoind with a profiler active to see if it is caused by a memory leak when the memory usage goes above 1Gb.

Looks to me your "friend" at "addr" : "xxx:3505" is missing about a 130,000 blocks ! (242438 - 114994). So I wouldn't call the 14Mb of data that you send to him BS. You're sending him the blockchain data from last year(s) ;-)

*grins* the perpetrator just revealed himself ;-)

Try to imagine you are a manufacturer of some hardware.
Can you imagine that you will need some hardware yourself to develop and maintain the software for your hardware ?
Well, that is test hardware.

That unit was a test machine shipped to a customer, a mistake, simple explanation.

Did you reset the Avalon to factory settings ?
Did you switch off the Avalon after resetting to factory default ?
Is your Avalon connected to a switch/router and your laptop connected to the same switch/router ?
Are you using the correct ethernet cables and not cross over cables ?
Are your laptop and Avalon on the same network segment ? 192.168.0.x
Are you running Windows ? If so, you'll have to reboot sometimes after changing network settings.
If you see a 169.x.x.x ip address, this could indicate a problem with your network card/cable. Reboot and try changing your settings again.
Is your netmask on your laptop set to 255.255.255.0 ?
Is your laptop ip address 192.168.0.10 ? (I use a fixed address here to keep it simple)
Can you ping the Avalon unit from your laptop ? ping 192.168.0.100
Is nmap http://nmap.org/download.html showing the Avalon on your network ?

Suppose there is a bug in handling the third and higher words of the target which are all 00000000 on testnet and not all 00000000 on main net ?
This would not be discovered when only running on test net.

And to answer your request for explanation:
If the machine generates a block and the network agrees with it, it is probably ok.
If the machine generates a share and the pool agrees with it, it is probably ok.
So feeding different input data is not an issue for Bitcoin hashing QA if you have a whole network validating your "test" result.

I would still also create a testset (and a program that processes the test data) with known input and output and nothing tells me Avalon doesn't have such a program and runs this test on every machine too before connecting to the main net for their final(?) test.

And maybe they also run against the test net as a second test. The value of this test will be limited, given the few machines running here, which is logical, testnet BTC have no value.

Without word from Avalon we're not sure what they do for testing, we can only speculate, just like what I am doing here.

Congrats! Really like to hear more about the process, but that's for another day!

Software developer here too.

Given the different difficulty of test net compared to main net, one has to run on the main net to be sure that the customer will not receive a malfunctioning unit. (And yes, I have looked at the Cgminer source code to see how the validation of a block works if the hash is calculated on a CPU, so I know what I am talking about).

I am sure Yifu will come up with a proper destination for the BTC generated, he is not in it for the money, remember ?

And to make the picture complete, I own Avalon units.

Avalon instructions say put the machine flat on the bottom.
The reason for this is that the heavy boards (about 10 lbs each) are only fixed with small screws to the bottom of the case.

Do you have any idea how many KILOWATTS you will need to run 50% of the Bitcoin network ?!
I'll help:
1 Avalon unit realizes 66Gh and uses 650W.
The network is 156418Gh, 50% is 78209Gh, so you will need 770 kilowatts of electricity to run those 1184 Avalon machines.

You will have 12 boards of 40W = 480W
Your power supply can deliver 70A @ 12V = 840W
80% optimum power drain = 480W, so 100%=600W in stead of 840W.
So you could go for a smaller PSU maybe ?

I have been searching on Google and the general consensus is that a 6pin PCI can easily handle 150W. The difference between the 75W 6 pin and 150W 8 pin is just 2 ground wires. Some people even put the 6pin cable into the 8 pin connector of a GPU without problems.

http://www.enterpoint.co.uk/cairnsmore/cairnsmore1.html

The Power Distribution Board could be used to split the ATX motherboard power connector into 6 PCI express connections.

Your solution indicates a much higher current than the 75W for a PCI cable. Is that OK ?

I was also busy with the power cables. It isn't that easy to find PCI splitter cables, but I found these: http://www.alibaba.com/product-gs/491429885/Y_splittler_power_extension_cable.html
minimum order 50 pieces.
Lets take 37.5W per board, so each 6 pin PCI power cable will handle 2 boards.
This will limit the number of boards per power supply, which only has a limited number of 6 pin PCI power connections.

Maybe it's better to find a power distribution board like Avalon is using, so we can split the power from the ATX motherboard connector into many 6 pin PCI cables ?
I'll continue searching.

Yifu said joking 'who says we're still in control of that address' ...
My guess, like stated in another thread is they sold the address to a big investor (Zuckerberg?) and made a deal for payments to that address...

Can you tell us more about this ?

Interesting read here: http://martinsliquidlab.i4memory.com/Radiator-Fan-Orientation-And-Shroud-Testing-Review.html
My gut feeling was wrong. Push + pull is better than just push. And adding two shrouds to it makes it even better.

Maybe MtGox is low on USD ?
Or in other words: more sellers doing USD withdrawal compared to buyers depositing USD...

I am surprised to see so many push pull fan setups in a closed case. If the pull fan sucks out more air than the push fan pushes in, the air pressure will be low in the case and there will be less air molecules to pickup and transport the heat.

I expect a push only fan setup to be much better, there will be lots of air molecules to pick up the heat and the air has no way to go except out.

Maybe even forcing a high pressure in the case by limiting the outgoing openings of the case. For example a 12cm fan blowing towards a 8x8 cm opening with smooth transition. I would like to use 12cm, so the board is 1cm away from the housing, creating some airflow over the remaining electronics on top of the board too.

My issue is it is a manual action. I want it to be automatic and limit the middle man attack risk.
Tyrion has stated he is in touch with WeExchange to see if it can become automatic, so let's wait and see.