I guess that the easiest way to compare both solutions (parallel vs sequential) is to show their performance through the elapsed time in computing the solution (once executed in the same machine). However, i think that other measurements would be interesting to have into account (such as memory, resources needed or another measure depending on your approach).

Any metric you take should ultimately be the reflection of how well or fast you solved the problem. So in your case, you can compare the metric against time taken to learn the solution. And as Martin pointed out, memory usage, computational cost etc., will be nice additions.