well, a small change might be something that requires huge amounts of memory, or even to use diskspace in order to be calculated, which makes the use of GPU so slow that CPU are still usable for mining. (take a look at the momentum algorithm used on protoshares, miners are only 2 or 3 times faster on GPU).
Requiring to use disc space is surely not a small change. Their
whitepaper seems to be outdated, when reading it the first thing I thought was that you can use bloom filter to significantly reduce memory requirements. Some googling revealed that someone else already pointed to this and other vulnerabilities in the proposed scheme were found. They seem to fixed them to some extent but its not fast to find any definite information. This all will require time, I can't just copypaste some code into SpreadCoin without understanding it.
GPU miner is my priority now.
Bloom filter is actually slower than the currently used solutions, since false positives are quite costly. Some people have tried it already, and using a regular hashtable and ignoring collisions is way faster (up to twice as fast depending on the implementation). But still, due to the size of the search space (2^50) and the amount of data to be hashed (2^26), even bloom filters would use an amount of memory that would not fit in the local memory of any GPU in the near future. So, even with a hashing data structure way more efficient than bloom filters (say, something twice as fast as the current implementation), we'd still have GPU and CPU mining ration bellow 5:1, which is quite acceptable. But then, momentum is not the only way to abuse the use of memory, there might be simpler ways to do so (preferably something that can be used with hashcash PoW, which makes for a much simpler transition). But then, i'm not the one deciding the priorities.
Yet, the fastest pts miner uses a kind of a bloom filter in LDS.