Of course, GPU miner was not ready on day 1. Brand new algo would take some R&D time. So far, we have an account of a single entity at a ~2x advantage (not exact figure: could be less, could be more). So essentially he has 2 CPU solo for every 1 CPU of joe public solo . And now the pools are gaining interest levelling the playing field.
I would not be surprised if a GPU miner was actually available on day one. Unfortunately (or possibly intentionally), the GPU implementation of Boolberry runs almost an order of magnitude faster on GPUs than CPUs on unoptimized code. Optimization was reportedly extremely easy for those who undertook it,
...Of course, the original CryptoNight implementation was crippled too without us realizing it, and we pushed the optimized code when it became available. But none of our core team had anything to do with that.
Elsewhere they say the difference in speed is 2.2 to 5 times as fast. That is far, far less then an order of magnitude.