OP proposal is not possible, even with 51% of the mining power behind it.
As dexX7 correctly pointed out, without knowing the private keys to the transactions, any transaction of the coins to a new input would invalidate the block (and any subsequent blocks). No client would accept this invalid chain, regardless of length.
I'm not sure why this proposal has cropped up multiple times a day since the mtgox incident. One almost suspects that there is an ongoing influx of new users who are unaware of the forum's search function.
K, any transactions based on the transactions of the gox BTC would be invalidated, but, any that weren't would just be just-as-valid.
Imagine we're on block 20, and, gox lost their coins on block 15, and, in block 15-20 five transactions occurred. Gox sent money from their controlled address to an unknown one (#16), the unknown address sent money to another unknown address (#17), some random user sent money to his friend (#18), his friend send money back to the random user (#19) and I donated some money to the EFF (#20).
If we go ahead and imagine all the miners said "Right, fuck everything after block fifteen, let's start again", they'd go back to block fifteen, GOX would have their coins, and, they'd just need to regenerate the blocks from there on, however, since they know about the five transactions, they could just pack it all into the first block, minus the gox ones, so:-
#16 is invalid, as, we don't want to do it (It's where gox loses their money)
#17 is invalid because it relies on #16
#18 is still valid, so, drop that in the first block
#19 is still valid, so, also drop that in the first block
#20 is still valid, so, also drop that final one into the first block.
Grand, now we're 16 blocks deep, and, we're still valid, with MTGox keeping their money. Now all we have to do is generate five more ($chainOne.length - $chainTwo.length + 1) blocks before the miners mining the first chain mine one, and, all nodes would jump to us as we're the longest chain, and, also a valid chain.
...at-least, that's my understanding.
It would be possible by creating a block number n and all miners agree to include this block. The only thing in this block would be a transaction to a new address and the destruction of the gox coins. So at an agreed time this would become the next block and miners would carry on mining the blocks after that. This is obviously only if they are lost and the addressees known.
Its a terrible idea for obvious reasons and any one thinking this would be a positive for bitcoin clearly do not understand it.
All miners is not sufficient to change the Bitcoin protocol it requires all users, all existing clients to be upgraded. Otherwise there is a fork and users will choose the one they want to use.
Users can choose, but, unless it faults the checkpoint system that bitcoin has in place (
here's a list), then, by default, bitcoin-qt/bitcoind will auto-jump to the longest, valid, chain. In this case, assuming miners can generate a longer, valid (But ignoring gox's transaction), chain before the miners on the current chain can extend it, then, we'd all auto-jump and orphan the other blocks.
EDIT:- Source to back me up:-
https://github.com/bitcoin/bitcoin/blob/f60e49d49c72908356d70d05ae30c6e63be2192d/src/main.cpp#L2001-L2005I admit, I didn't just read the entire bitcoin source, so, I'm going mainly off comments & function names.