the technical term for a mutational event that undo the effect of a mutation is reversion.
Most beneficial mutation are not reversion events but reversion events are usually consider beneficial.
You have to keep in mind that if most beneficial mutation were reversion, nothing would really change and everything would be stagnating between two different "condition/state: the mutated state and the non-mutated state.
Generalizing a bit: if your statement were true, then reversions would have a distinct selective advantage, and reversions would win out, except in the case of a rapidly changing environment where a 'new' gene may have even greater selective advantage. So we would see 'punctual' evolution completely predominating - ie., we only get new species when environments change drastically. But. We see new species occuring in the fossil record pretty much continuously.
The point is that if reversion events were highly selected for, we'd have almost no continuing speciation.
You should see it the other way around. For a most good mutations to be a reversion mutation (according to kmarinas), it would mean that the bad mutation had been selected for in the population (otherwise the majority of the population already has the good allele).. which is unlikely (although there is genetic drift, bottlenecks, or tight linkage with a good allele).
I'm not sure about jim mcnamara's point, about reversions winning out.. what are the chances of lightening striking the same spot twice? It's the good allele that will win out over the bad mutation.