Indeed, it's trivially possible. You don't change them after the fact, you change them before you mine. P2Pool enables this, and any pool that supports GBT (e.g. Eligius) could in theory support this. (I believe eligius actually does, but I don't think any miner software exists that lets you make use of it).
Selecting transactions is really not at all sufficient, as it would still let someone controlling a pool reverse transactions. What you really want to do is separate "pooling for payment" from "control of the Bitcoin consensus". This is trivially accomplished, from a technical perspective: Do a GBT request to a pool to get a coinbase transaction that they'll accept, then query a local bitcoin daemon for your idea of the network consensus... merge the two and submit shares to the pool. The pool then pays you if your shares are paying a suitable (e.g. winnings go to the pool) coinbase. ... but this requires non-existing miner software and, of course, pools which support pooling only for the payment.
You would think that these supposedly civic minded pool operators would have getting themselves out of the creators-of-systemic-risk business as a high priority, but apparently not, since the market apparently doesn't demand it. (Eligius has done the most in this regard for centralized pools ... and it's 2.5% of the hashrate. P2Pool is a decentralized system and it's about 1% of the hashrate).
At least the hardware (AMD, NVidia GPU) is within reach of most people that want to mine Litecoin.
Litecoin was originally marketed as GPU proof, as you can see that didn't pan out so well. Many people believe that people were secretly GPU mining it all along while others druged along with CPUs. Likewise, there may be people with FPGA or ASIC miners on LTC today. There is nothing fundamental that ltc is doing which makes it hard to get a large advantage... maybe not quite as much of an advantage, but a 10x advantage in a thin margin business might as well be infinite.
